electronics Article
Methane Gas Density Monitoring and Predicting Based on RFID Sensor Tag and CNN Algorithm Chunlei Zhang 1 , Yuhua Fu 2, *, Fangming Deng 3, *, Baoquan Wei 3 and Xiang Wu 3 1 2 3
*
School of Resource and Environmental Engineering, Jiangxi University of Science and Technology, Ganzhou 341000, China;
[email protected] College of Applied Science, Jiangxi University of Science and Technology, Ganzhou 341000, China School of Electrical and Automation Engineering, East China Jiaotong University, Nanchang 330013, China;
[email protected] (B.W.);
[email protected] (X.W.) Correspondence:
[email protected] (Y.F.);
[email protected] (F.D.); Tel.: +86-797-8312-759 (Y.F.); +86-791-8704-6203 (F.D.)
Received: 9 April 2018; Accepted: 10 May 2018; Published: 12 May 2018
Abstract: According to the advantages of integrating wireless sensors networks (WSN) and radio frequency identification (RFID), this paper proposes a novel method for methane gas density monitoring and predicting based on a passive RFID sensor tag and a convolutional neural networks (CNN) algorithm. The proposed wireless sensor is based on electronic product code (EPC) generation2 (G2) protocol and the sensor data is embedded into the identification (ID) information of the RFID chip. The wireless sensor consists of a communication section, radio-frequency (RF) front-end section, and digital section. The communication section is used to perform the transmission and reception of wireless signals, modulation, and demodulation. The RF front-end section is adopted to provide the stable supply voltage for other parts. The digital section is employed to achieve sensor data and control the overall operation of the wireless sensor based on EPC protocol. Because the miscellaneous noises will decrease the accuracy during the process of data wireless transmission, the CNN algorithm is adopted to extract the robust feature from raw data. The measurement results show that the exploited RFID sensor can realize a maximum communication distance of 10.3 m and can accurately measure and predict the methane gas density in an underground mine. The RFID sensor technology is a beneficial supplement to the current underground WSN monitoring system. Keywords: methane gas density; RFID sensor tag; CNN algorithm
1. Introduction Because of the existence of methane and other toxic gases, underground mines have proved to be dangerous [1–3]. For example, 33.8% of accidents in underground mines are caused by methane gas density exceeding the normal range [4]. Generally speaking, these underground mine accidents are difficult to eliminate because human beings always work underground. In all accidents, methane exposure is the most common and the most serious [5,6]. The minimum explosion percentage of methane gas is 5%. However, when exposed to low methane gas density for a long time, the human body will still be severely damaged [7–9]. The methane gas density is traditionally measured by human beings [1,3]; however, this method is not only time-consuming and laborious, but dangerous to these people in underground mines. A search-and-rescue robot system was introduced for the remote sensing of the underground coal mine environment [10]. It can achieve environment sensing and rescue automatically. However, it is still not a real-time monitoring system. In the past few decades, the cable monitoring system (CMS) has been widely employed in underground mines for real-time environment monitoring [2]. Still, due to
Electronics 2018, 7, 69; doi:10.3390/electronics7050069
www.mdpi.com/journal/electronics
Electronics 2018, 7, 69
2 of 13
the complex situation of underground mines, the CMS requires high installation and maintenance costs. Therefore, the environmental monitoring of underground mines urgently needs a flexible and economic method. In recent years, due to their advantages of fast deployment and low cost, wireless sensor network (WSN) technologies have been wildly applied in industry [11–13]. In particular, ZigBee technology has become the mainstream technology for underground environment monitoring [14–16]. Nevertheless, the WSN network does not care about the location of the sensor node; so, when an abnormal situation happens, the abnormal node cannot be identified quickly. Hence, the research on integrating WSN with radio frequency identification (RFID) arouses great interest [17–19]. Each RFID tag has a unique identification (ID) number. The integration of these two technologies is beneficial for the WSN network to concentrate on the data transmission. When the information of a specific node is considered, this node can be quickly located by using the RFID function. Furthermore, because the RFID sensor employs a backscattering communication scheme, it can work in passive mode for long-term application. Hence, RFID sensor technology is a beneficial supplement to the current WSN technology. While there are always several research projects focusing on the combination of RFID and WSN technology [20–22], this paper studied the feasibility of monitoring methane gas density in an underground mine environment by using a wireless sensor based on RFID technology for the first time. RFID sensors can be classified into two categories, the chipless group [17,23] and chip-based group. Chipless RFID sensors adopt a sensor’s signal to directly modulate the backscattering performances of the antenna. They have the advantage of being ultra-low cost; however, they are not suitable for composing the sensor network due to the lack of digital blocks. Chip-based RFID sensors can also be divided into two groups, integrated circuit (IC)-based RFID sensors [18,19] and printed circuit board (PCB)-based RFID sensors [24,25]. IC-based RFID sensors integrate the sensors with an RFID tag chip, resulting in small circuit area and high stability. However, these sensors require a long design period. Moreover, only several kinds of sensor can be fabricated by the complementary metal oxide semiconductor (CMOS) process. PCB-based RFID sensors incorporate an RFID chip, sensors, and digital blocks on the same board. These sensors show the advantages of short design period, high extension capability, and low cost. Besides real-time monitoring, gas density prediction is also important for mine safety. Due to the harsh underground environment, the data measured by the passive wireless sensor contains miscellaneous noises, which makes it difficult to extract robust features from the raw signal by using traditional methods [26–28]. In recent years, deep learning technology with autonomous learning ability has been widely researched and has made surprising progress in feature learning [29–32]. Deep belief networks (DBNs) [33], stacked denoising auto-encoders (SDAs) [32], and convolutional neural networks (CNNs) [34] are three widely employed deep learning approaches. CNN shows priority in feature extraction and data compression [35]. Therefore, in this paper CNN is employed for feature extraction. This paper presents a novel wireless sensor based on RFID technology for methane gas monitoring in the long term. It is a beneficial supplement to the current underground WSN monitoring system. Considering the environment interruption on data wireless transmission, CNN is adopted to extract robust and discriminative features from the raw signal. 2. RFID Sensor Design 2.1. Communication Mechanism The RFID system generally consists of an RFID tag and an RFID reader. Figure 1 shows the basic communication flow based on the electronic product code (EPC) generation2 (G2) protocol [36]. Within the communication distance, the reader first sends the Select instruction by carrier wave (CW) to the tag to change the tag’s status. The reader then sends the Query instruction to the tag and the tag
Electronics 2018, 7, 69 Electronics 2018, 7, x FOR PEER REVIEW
3 of 13 3 of 12
information CRC code thethe tag. Finally, Req_RN is instruction sent by reader to acquire the Handle responds withand its RN16. Afterof that, reader sends the ACK to obtain the ID information response of theoftag. and CRC code the tag. Finally, Req_RN is sent by reader to acquire the Handle response of the tag. 3Req_RN of 12
CW
Req_RN
Handle
EPC+CRC16
RN16
ACK
Send instruction
CW
Send instruction
Initialize
Query
Random number
Send Send instruction
RFID tag
CW
Select
Send instruction
CW
Reader
EPC+CRC16
information and CRC code of the tag. Finally, Req_RN is sent by reader to acquire the Handle response of the tag.
confirmation
Handle
Random number
CW
ACK
Send instruction
CW
Query
Send instruction
CW
instruction
Send instruction
CW Reader Select Electronics 2018, 7, x FOR PEER REVIEW
release
Figure 1. 1. Schematic Schematic diagram diagram of of radio radio frequency frequency communication. communication. Figure RFID tag
Initialize
RN16
confirmation
release
From the information discussed above, we found that there is no direct instruction for processing From the information discussed above, we found there is no direct instruction for processing Figure 1. Schematic diagram of radio that frequency communication. sensor information; hence, we have to adjust this communication flow. In general, the sensor data are sensor information; hence, we have to adjust this communication flow. In general, the sensor data are stored in the non-volatile memory of the tag we chip, butthat it there exhibits long instruction responsefor time and high power From the information discussed above, found is no direct processing stored in the non-volatile memory of the tag chip, but it exhibits long response time and high power sensor information; hence, we have to adjust this Thus, communication flow. Inthe general, the sensor data are consumption during the read/write process [20]. in this paper sensor data is embedded into consumption during the read/write process Thus, in this paper thetime sensor data is embedded stored in the non-volatile memory theRFID tag [20]. chip, exhibits long response and high power the identification (ID) information of of the tag.butInitthis way, the time for reading and writing the into the identification (ID) information of the[20]. RFID tag. In this way, the time for readinginto and writing consumption during the read/write process Thus, in this paper the sensor data is embedded sensor data into the non-volatile memory can betag. avoided, resulting in lower power consumption and identification (ID) information of the RFID this way, the time for reading andpower writingconsumption the the sensorthe data into the non-volatile memory can beInavoided, resulting in lower response sensor time data compared with the traditional method. Figure 2 shows the comparison of the ID into the non-volatile memory can be avoided, resulting in lower power consumption and and response time compared with the traditional method. Figure 2 shows the comparison of the ID response time the compared with the traditional Figure 2 shows the comparison of the ID information between traditional method andmethod. the proposed method. information betweenbetween the traditional method proposed method. information the traditional methodand and the the proposed method.
Conventional EPC organization Conventional EPC organization ID ID 5-bit5-bit
NULL NULL
CRCCRC 16-bit 16-bit
(a)
(a)
Proposed ID organization
Proposed ID organization Sensor Sensor Sensor Version
Sensor type CRC ... data2 data1 Sensor dataN 16-bit 8-bit Sensor 8-bit Sensor type ... Sensor Version data1 data2 (b) dataN
8-bit
8-bit
CRC 16-bit
Figure 2. Comparison of sensor data storage method; (a) traditional method; (b) proposed method.
(b) (a) traditional method; (b) proposed method. Figure 2. Comparison of sensor data storage method; Anti-collision technology is common in the field of RFID systems. However, in the field of environment monitoring ofisunderground position of the RFID sensor tag is fixed. The Anti-collision technology common mines, in thethe field of RFID systems. However, in the field number of the RFID sensor tag and the density of the RFID reader are not as much as in the field of of environment monitoring of underground mines, the position of the RFID sensor tag fixed. Anti-collision is common the field of RFID However, the isfield of supply chaintechnology or health monitoring. Henceinthe anti-collision problemsystems. is not a serious issue ininthe The number of the RFID sensor tag and the density of the RFID reader are not as much as in the application of underground mine monitoring. As for ourposition design, weof propose the Q algorithm in EPC environment monitoring of underground mines, the the RFID sensor tag is fixed.field The protocol [36]monitoring. for tagHence anti-collision andRFID the reader-coverage avoidance of supplyofGen2 chain or health theofanti-collision problem iscollision not a serious in the number the RFID sensor tagRFID and the density the reader are not as much as inissue the field of arrangement (RCCAA)mine method [37,38] for RFID anti-collision. application of underground monitoring. Asreader for our design, we propose the Q algorithm in EPC
Figure 2. Comparison of sensor data storage method; (a) traditional method; (b) proposed method.
supply chain or health monitoring. Hence the anti-collision problem is not a serious issue in the Gen2 protocol [36] for RFID tag anti-collision and arrangement application underground mine monitoring. Asthe forreader-coverage our design, we collision propose avoidance the Q algorithm in EPC 2.2.of Wireless Sensor Design (RCCAA) method [37,38] for RFID reader anti-collision. Gen2 protocol [36] for RFID tag anti-collision and the reader-coverage collision avoidance The block diagram of the proposed wireless sensor is illustrated in Figure 3a. It consists of three
arrangement method [37,38] forfront-end RFID reader parts:(RCCAA) a communication section, an RF section, anti-collision. and a digital section. The communication 2.2. Wireless Sensor Design section contains the antenna and RFID tag chip, which are employed to perform the transmission and reception ofDesign wireless signals, modulation, and demodulation. The RF front-end section, composed of 2.2. Wireless Sensor The block diagram of the proposed wireless sensor is illustrated in Figure 3a. It consists of three a matching network, a four-stage rectifier, and a low dropout (LDO) voltage regulator, is adopted to parts:The a communication section, an RF section, a digital The communication providediagram the stableof supply voltage forfront-end other parts. sensor The matching network, containing RFItinductor block the proposed wireless isand illustrated insection. Figurean3a. consists of three section contains antenna and RFID tag chip, which employed to perform the transmission and (Lm) andthe an adjustable ceramic trimmer capacitor (Cm), are is employed to realize the maximum parts: a communication section, an RF front-end section, and a digital section. The energy communication reception of wireless signals, modulation, demodulation. The RF front-end section, composed of a section contains the antenna and RFID tagand chip, which are employed to perform the transmission and matching network, a four-stage rectifier, and a low dropout (LDO) voltage regulator, is adopted to reception of wireless signals, modulation, and demodulation. The RF front-end section, composed of provide the network, stable supply voltage rectifier, for other and parts. Thedropout matching network, containing anisRF inductor a matching a four-stage a low (LDO) voltage regulator, adopted to (L an adjustable ceramic trimmer capacitor (Cm ),matching is employed to realize the maximum energy m ) and the provide stable supply voltage for other parts. The network, containing an RF inductor
(Lm) and an adjustable ceramic trimmer capacitor (Cm), is employed to realize the maximum energy
Electronics 2018, 7, 69 Electronics 2018, 7, x FOR PEER REVIEW Electronics 2018, 7, x FOR PEER REVIEW
4 of 13 4 of 12 4 of 12
transferring transferring efficiency. efficiency. The The power power harvested harvested by by the the antenna antenna is is an an ultra-high ultra-high frequency frequency signal, signal, which which transferring efficiency. The power harvested antenna is directly. an ultra-high frequency signal, which cannot used power the remaining part ofthe the RFID sensor directly. Therefore, as shown in cannot be used totopower the remaining part ofby the RFID sensor Therefore, as shown in Figure cannot be used to power the remaining part of the RFID sensor directly. Therefore, as shown in Figure Figure 3b, a four-stage rectifier is introduced to rectify and the boost the harvested the required 3b, a four-stage rectifier is introduced to rectify and boost harvested powerpower to the to required direct 3b, a four-stage rectifier is introduced to rectify and is boost harvested power toathe required direct direct current (DC) voltage. The LDO voltage regulator isthe then used to generate stable voltage current (DC) voltage. The LDO voltage regulator then used to generate a stable DC DC voltage for current (DC) voltage. The LDO voltage regulator is then used to generate a stable DC voltage for for other parts. The digital section, consisting of a micro-controller unit (MCU) and a methane gas other parts. The digital section, consisting of a micro-controller unit (MCU) and a gas other parts. Theis section, consisting ofdata a micro-controller (MCU) and aof gas density sensor, employed to sensor and overall operation the density sensor, isdigital employed to achieve achieve sensor data and control control the theunit overall operation ofmethane the wireless wireless density sensor, employed to achieve sensor data and control the overall operation of the wireless sensor on G2 sensor based based onisEPC EPC G2 protocol. protocol. sensor based on EPC G2 protocol. RF front-end RF front-end Lm Lm Cm
rectifier
C3 DC Voltage Digital section
C2
section
Digital section
C2
2
I C bus
chip
MCU MCU
Methane gas density Methane ADC gassensor density ADC
sensor
V1 V1
C1 C1 D1 D1
D6 D5
C3
Communication section Communication
D8
D7
D6
rectifier
DC Voltage
I2C bus
D7
Four stage rectifier Four stage
C4
regulator
Cm
RFID tag chiptag RFID
D8
C4
LDO voltage LDO regulator voltage
Four stage rectifier Four stage
D4
D5
D4 D3
D2
D3
D2 C5
C6
C7
C8
C9
C10
C11
C5
C6
C7
C8
C9
C10
C11
(a) (a)
(b) (b)
Figure 3.3.Proposed Proposedradio radio frequency identification (RFID) sensor tag:architecture (a) architecture of thesensor RFID Figure frequency identification (RFID) sensor tag: (a) of the RFID Figure 3. Proposed radiofour-stage frequencyrectifier. identification (RFID) sensor tag: (a) architecture of the RFID sensor tag; (b) proposed tag; (b) proposed four-stage rectifier. sensor tag; (b) proposed four-stage rectifier.
Figure 4 shows the prototype of the exploited RFID sensor tag. Aiming at the low-cost Figure 44shows thethe prototype of the RFID sensor Aiming the low-cost application, Figure shows prototype ofexploited theboth exploited RFID tag. sensor tag. atAiming at the application, a dipole antenna is adopted for communicating and energy harvesting. Thelow-cost antenna a dipole antenna is adopted for both communicating and energy harvesting. The antenna isantenna etched application, a dipole antenna is adopted for both communicating and energy harvesting. The is etched on the FR4 substrate, the dielectric constant and thickness of which are 4.5 and 1.5 mm, on the FR4 substrate, the dielectric constant and thickness of which are 4.5 and 1.5 mm, respectively. is etched on the substrate, the is dielectric constant and are 4.5 and 1.5Ω.mm, respectively. TheFR4 RFID chip chosen Monza X-8K, with an thickness impedanceofatwhich 915 MHz of 19–172 The The RFID chipThe chosen ischip Monza X-8K,iswith an impedance at 915 MHz of 19–172 Ω. The rectifier adopts respectively. RFID chosen Monza X-8K, with an impedance at 915 MHz of 19–172 Ω. The rectifier adopts a zero-bias Schottky diode SMS7630 as the switch diode for high rectifying efficiency. a zero-bias Schottky diode Schottky SMS7630 diode as theSMS7630 switch diode forswitch high rectifying efficiency. The NCP583 is rectifier adopts a zero-bias aswhich the diode for1.8 high rectifying efficiency. The NCP583 is selected as the LDO voltage regulator, can generate V output voltage within selected as the voltage regulator, whichregulator, can generate 1.8can V output voltage the inputwithin range The NCP583 isLDO selected as the voltage which generate 1.8 V within output voltage the input range from 1.65 to LDO 1.83 V. The MCU adopts MSP430FR6964, which can provide 2526 kB from 1.65 to 1.83 V. The MCU adopts MSP430FR6964, which can provide 2526 kB FRAM and a 12-bit the input range from 200-ksps 1.65 to 1.83 V. The MCU adopts MSP430FR6964, provide 2526 118 kB FRAM and a 12-bit Analog-to-Digital Converter (ADC). Thiswhich MCUcan only consumes 200-kspsand Analog-to-Digital Converter (ADC). ThisConverter MCU only(ADC). consumes 118MCU µA/MHz at 1.8 V supply FRAM a 12-bit 200-ksps Analog-to-Digital This only consumes 118 μA/MHz at 1.8 V supply voltage in the measurement mode. The employed methane gas density voltage inatthe mode.inThe methane gasThe density sensormethane is MQ-4,gas which can μA/MHz 1.8measurement V supply theemployed measurement mode. density sensor is MQ-4, which canvoltage detect the methane gas density within theemployed range of 300–10,000 ppm. detect the methane gas density within the range of 300–10,000 ppm. sensor is MQ-4, which can detect the methane gas density within the range of 300–10,000 ppm.
Antenna Antenna RFID tag chip RFID tag chip Matching network and Matching rectifier network and rectifier
Methane gas Methanesensor gas density density sensor
MCU MCU
Figure 4. The introduced four-stage rectifier. Figure Figure 4. 4. The The introduced introduced four-stage four-stage rectifier. rectifier.
Figure 5 shows the whole procedure of monitoring and predicting methane gas density. The Figure 5 shows the tag whole procedure monitoring and predicting methane The proposed RFID sensor is adopted to of collect the methane gas density data, gas and density. the CNN is proposed RFID sensor tag is adopted to collect the methane gas density data, and the CNN is employed to extract robust features from the raw data. After that, the least squares support vector employed to extract robust features from the raw data. After that, the least squares support vector
Electronics 2018, 7, 69
5 of 13
Figure 5 shows the whole procedure of monitoring and predicting methane gas density. 5 of 12 adopted to collect the methane gas density data, and the CNN 5 of 12 is employed to extract robust features from the raw data. After that, the least squares support vector regression (LS-SVR) is based on the extracted features. Then, LS-SVR regression is established established Then, the the established established LS-SVR is is regression (LS-SVR) (LS-SVR) is established based based on on the the extracted extracted features. features. Then, the established LS-SVR is used to predict the methane gas density. used to predict the methane gas density. density. Electronics 2018, 7, xRFID FOR PEER REVIEW The proposed sensor tag is Electronics 2018, 7, x FOR PEER REVIEW
Collecting data by Collecting data by RFID sensor RFID sensor
CNN feature CNN feature extraction extraction
Robust features Robust features
Subsequent Subsequent process process
LS-SVR LS-SVR methane gas methane gas density prognosis density prognosis
LS-SVR LS-SVR Modeling Modeling
Figure 5. The whole procedure of monitoring and predicting the methane gas density. Figure 5. The whole procedure of monitoring and predicting the methane gas density. Figure 5. The whole procedure of monitoring and predicting the methane gas density.
3. Proposed Algorithm 3. Proposed Algorithm 3. Proposed Algorithm 3.1. Convolutional Neural Network 3.1. Convolutional 3.1. Convolutional Neural Neural Network Network The basic structure of the CNN is shown in Figure 6. Each sample data is input in twoThe basic structure the CNN is shown in Figure 6. Each data sample datain is input in twoThe basic structure of of thethe CNN is shown 6. Each sample is input two-dimensional dimensional form, and then hidden layerinisFigure input by kernel function. The hidden layer is mainly dimensional form, thenlayer the hidden is input by kernel function. The hidden layercomposed is mainly form, and then theand hidden inputlayer by kernel hidden layer is mainly composed of alternately repeatedisconvolution layerfunction. and poolThe layer, and the output of this layer is composed of alternately repeated convolution layer and pool layer, and the output of this layer as is of alternately convolution layerhas anda pool layer, and the output of input this layer is sampled sampled as therepeated next layer. The structure higher fault tolerance to the samples, and can sampled as the The nextstructure layer. The structure has a higher faulttotolerance the input samples, and can the next layer. has aofhigher fault tolerance theconvolution inputtosamples, and can realize the realize the hierarchical expression data more accurately. The layer is used to extract realize the hierarchical expression of data more accurately. The convolution layer is used to extract hierarchical expression of data more accurately. The convolution layer is used to extract the local the local features of input data and consists of multiple feature matrices. Each characteristic matrix the local of features of input data and of consists of multiple feature matrices. Each characteristic matrix features input data consists multiple feature kernel matrices. characteristic matrix canthe be can be regarded as a and plane (the same convolution on Each the same plane), so it has can be regarded as a plane (the same convolution kernel on the same plane), so it has the regarded as a plane (the same convolution kernel on the same plane), so hasparameters. the characteristics of characteristics of parallel computation and greatly reduces the number of itfree Different characteristics of parallel computation and greatly reduces the number ofDifferent free parameters. Different parallel computation and greatly reduces the number of free parameters. planes correspond planes correspond to different convolution kernels so that the extracted features are more fully planes correspond to different convolution kernelsfeatures so that are themore extracted features are more fully to different convolution kernels so that the extracted fully demonstrated. demonstrated. demonstrated.
Output Output Input Input Convolutional Convolutional nuclei nuclei
C1 C1
S1 S1
C2 C2
S2 S2
Unfolded Unfolded
Full Full connection connection
Figure Figure 6. 6. Basic Basic structure structure of of the the convolution convolution neural neural network. network. Figure 6. Basic structure of the convolution neural network.
The calculation process of the convolution layer is shown in Figure 7a. The calculation method calculation process process of ofthe theconvolution convolutionlayer layerisisshown shownininFigure Figure7a. 7a.The Thecalculation calculation method The calculation method is is as follows: is as follows: as follows: l −1 l l l f l 1 l l Y lY= X × K + B (1) f ∑ X l 1 K l B l (1) Y l f X K B (1) In the formula, l represents the number of layers, f is the activation function, K is the convolution In the formula, l represents the number of layers, f is the activation function, K is the convolution kernel, B isformula, the bias, lXrepresents is the input andofYlayers, is the output value. In the thevalue, number f is the activation function, K is the convolution kernel, B is the bias, X is the input value, and Y is the output value. kernel, B is the bias, X is the input value, and Y is the output value. The pool layer is used to compress data on the results of the stacked layer to reduce data The pool layer is used to compress data on the results of the stacked layer to reduce data dimensions. The feature extracted by the pool layer has no deformation at scale and prevents over dimensions. The feature extracted by the pool layer has no deformation at scale and prevents over fitting. The pooling process is shown in Figure 7b, and the calculation is as follows: fitting. The pooling process is shown in Figure 7b, and the calculation is as follows: 1 Y ll f 1 X ll B ll (2) Y f k X B (2) k A fully-connected layer is connected after the combination of some convolutional and pooling A fully-connected layer is connected after the combination of some convolutional and pooling layers. The fully-connected layer shows similar function to the traditional multilayer neural network layers. The fully-connected layer shows similar function to the traditional multilayer neural network
Electronics 2018, 7, 69
6 of 13
The pool layer is used to compress data on the results of the stacked layer to reduce data dimensions. The feature extracted by the pool layer has no deformation at scale and prevents over fitting. The pooling process is shown in Figure 7b, and the calculation is as follows: Yl = f
1 X l + Bl k∑
(2)
Electronics 2018, 7, x FOR PEER REVIEW A fully-connected layer is connected
6 of 12 after the combination of some convolutional and pooling layers. The fully-connected layer shows similar function to the traditional multilayer neural network and can be beapplied appliedthrough through different classification models. The softmax regression, which can and can different classification models. The softmax regression, which can achieve achieve fast computation and an accurate is the most popular output layer. Theof output of the fast computation and an accurate result, isresult, the most popular output layer. The output the softmax softmax regression can be obtained as follows: regression can be obtained as follows:
exp X K B 1 1 exp( X × K1 + B1 ) + B2B 2 ) 1 1 2 2 exp(XX×KK exp O =O n n ... . . . expX × X B exp KK + B ∑ j j j j exp(XX×KK + Bn ) exp j=1 j 1 n B n n
(3) (3)
Figure diagram of of thethe convolution andand pooling process: (a) convolutional process Figure 7. 7.Schematic Schematic diagram convolution pooling process: (a) convolutional process schematic; (b) pooling process schematic. schematic;
In the learning process, the back-propagation algorithm is adopted, that is, the weight matrix is In the learning process, the back-propagation algorithm is adopted, that is, the weight matrix is adjusted by reducing the mean square error (MSE) of the ideal output and the actual output. The adjusted by reducing the mean square error (MSE) of the ideal output and the actual output. The MSE MSE is calculated as follows: is calculated as follows: 1 1 ∑ y j − o j 22 (4) MSE = MSE 2 y j o j (4) j 2 j
Here, y is the actual output and o is the ideal output. Here, yjj is the actual output and ojj is the ideal output. The essence of the convolution neural network is to learn a number of filters that can extract The essence of the convolution neural network is to learn a number of filters that can extract the the characteristics of the input data, extract the topological features hidden in the data through characteristics of the input data, extract the topological features hidden in the data through layer-bylayer-by-layer convolution and pooling, and finally obtain the input data with translation, rotation, layer convolution and pooling, and finally obtain the input data with translation, rotation, and scaling, and scaling, the characteristics of the nature of change. This method can learn the features implicitly the characteristics of the nature of change. This method can learn the features implicitly from the data, from the data, avoid explicit feature extraction, and achieve higher accuracy and efficiency than avoid explicit feature extraction, and achieve higher accuracy and efficiency than traditional neural traditional neural networks. networks. 3.2. Least Squares Support Vector Regression For a training dataset {(xi, yi)|i = 1, 2, …, n}, xi R N , yi R , (n is the elements number in the data set and N is the dimension of x), the LS-SVR model is defined as [39]:
Electronics 2018, 7, 69
7 of 13
3.2. Least Squares Support Vector Regression For a training dataset {(xi , yi )|i = 1, 2, . . . , n}, xi ∈ R N , yi ∈ R, (n is the elements number in the data set and N is the dimension of x), the LS-SVR model is defined as [39]: f ( x ) = wT · ϕ( x ) + b
(5)
where w is the coefficient matrix, b is the bias term, and ϕ(x) is employed to project the data x into a higher dimensional space. The LS-SVR is determined by w and b, so the target is to search for the parameters which can acquire a satisfactory generalization performance. This target can be achieved by minimizing: 2 1 2 kwk
+
C 2
s.t. yi = w T x + b + ei ,
n
∑ ei2
(6)
i =1
i = 1, 2, · · · , n
where ei is used to minimize the deviations of the regression function, and C is used to control the compromise between the complexity and the fitness to the data. The Lagrange multiplier approach is employed to solve the optimization problem: "
b α
#
"
=
1T K + I/c
0 1
# −1 "
0 y
# (7)
where 1 = [1, 1, . . . , 1], α = [α1 , α2 , . . . , αn ]T , y = [y1 , y2 , . . . , yn ]T , and K = k(xi , xj ) is the kernel function. It can be obtained by the kernel trick: k(xi , xj ) = ϕ(xi )× (xj ). Finally, the LS-SVR model is equivalent to [36]: n
f (x) =
∑ αi k ( xi , x ) + b
(8)
i =1
The kernel function significantly influences the performance of the LS-SVR model. Since the raw signal is characterized as non-linear and non-stationary, the radial basis function (RBF) is used as the kernel function [40]: n o K ( xi , x ) = exp −k x − xi k2 /σ2 (9) where σ is the bandwidth of the kernel. For a series of observations Xt = [xt −(n−1)s , xt −(n−2)s , . . . , xt ], s is the interval of measurement and n is the series length. The goal of prediction is to estimate xt+r based on Xt . The estimated value can be calculated by: ∧
xt+r = f ( xt , xt− p , xt−2p , · · · , xt−np )
(10)
∧
where xt+r is the estimated value, and f (x) is the trained LS-SVR. 4. Experimental Results and Discussion In this paper, compared with the conventional sensor data stored in non-volatile memory (NVM), the identification (ID) organization is optimized to achieve less communication time delay and longer communication distance. Therefore, the time delay and maximum communication distance of both the traditional method and the proposed method were experimentally measured. The exploited wireless sensor operates at 915 MHz, and the radiation power of the reader is 4 W. The operating distance between the wireless sensor and the reader was set to 1.5 m. The measured results are listed in Table 1.
Electronics 2018, 7, 69
8 of 13
Table 1. Time delay comparison. Method
Data Location
Operation
Time (ms)
Traditional Proposed
NVM ID
Inventory and Read Inventory
7.45 ± 3.92 22.13 ± 3.31
Since the supply voltage to the sensor tag is fixed at 1.8 V, the whole power consumption of the RFID sensor tag is determined by the consumed current. The proposed sensor is powered by a DC power supply Yokohama, Japan), and the consumed current is measured8by a Electronics 2018, 7, PMX-35A x FOR PEER (Kikusui, REVIEW of 12 digital multimeter Keithley 2010 (Tektronix, Shanghai, China). The developed sensor tag is activated every 30 ms by an Figure 8 illustrates the current waves of the RFID employing activated every 30 RFID ms byreader. an RFID reader. Figure 8 illustrates the current waves of sensor, the RFID sensor, these two different sensor data storage methods. The measured currents of the proposed method and employing these two different sensor data storage methods. The measured currents of the proposed the traditional method are 207 µA and µA, proposed sensor tag still has a current method and the traditional method are478 207 μArespectively. and 478 μA,The respectively. The proposed sensor tag consumption of 4.2 µA in sleep mode. still has a current consumption of 4.2 μA in sleep mode.
Figure 8. Comparison of the power consumption under two different sensor data storage Figure 8. Comparison of the power consumption under two different sensor data storage mechanisms. mechanisms.
The received signal strength indication (RSSI)(RSSI) is one is aspect which evaluate The intensity intensityofofthethe received signal strength indication one with aspect withtowhich to the performance of the proposed wireless sensor. The RSSI can be calculated by: evaluate the performance of the proposed wireless sensor. The RSSI can be calculated by:
=10log YY 10 logXX RR=
(11) (11)
where Y YR is the RSSI (dBm) and X is the received signal power (mW). As shown in Figure 9a, the RSSI where R is the RSSI (dBm) and X is the received signal power (mW). As shown in Figure 9a, the RSSI performance proposed wireless sensor was performed in the underground mine. The performance ofofthethe proposed wireless sensor was performed in the underground mine. The experiment experiment results areinillustrated Figure 9b. Itthat canwithin be seen that withinofthe distance of 10.3 m, the results are illustrated Figure 9b. in It can be seen the distance 10.3 m, the RSSI is higher RSSI is higher than −60 dBm. We know from Reference [41] that this result is acceptable. than −60 dBm. We know from Reference [41] that this result is acceptable. The proposed RFID communication protocol requires a bit error rate (BER) less than 0.001 and Wireless sensor signal to noise ratio (SNR) higher than 25 dB [36]. The BER performance is determined by the communication environment and SNR, therefore the SNR over distance and the BER under different SNRs should be measured. Figure 10a shows the measured SNR under different distances; it can be seen that within the communication distance of 10.3 m, the measured SNR is higher than 26 dB. Tester Figure 10b shows the measured BER under different SNR. As seen in the figure, when the SNR is higher than 25 dB, the measured BER is less than 0.001. Therefore, the proposed wireless sensor works well in the test scenario. (a)
(b)
Figure 9. (a) Received signal strength indication (RSSI) measurement environment; (b) measurement results. The proposed RFID communication protocol requires a bit error rate (BER) less than 0.001 and signal to noise ratio (SNR) higher than 25 dB [36]. The BER performance is determined by the communication environment and SNR, therefore the SNR over distance and the BER under different
YR =10log X
(11)
where YR is the RSSI (dBm) and X is the received signal power (mW). As shown in Figure 9a, the RSSI performance of the proposed wireless sensor was performed in the underground mine. The Electronics 2018, 7, 69 9 of 13 experiment results are illustrated in Figure 9b. It can be seen that within the distance of 10.3 m, the RSSI is higher than −60 dBm. We know from Reference [41] that this result is acceptable. Wireless sensor
Tester
(a)
(b)
Figure 9. (a) Received signal strength indication (RSSI) measurement environment; (b) Figure 9. (a) Received signal strength indication (RSSI) measurement environment; measurement results.
(b) measurement Electronics 2018, 7, x FOR results. PEER REVIEW
9 of 12
The proposed RFID communication protocol requires a bit error rate (BER) less than 0.001 and signal to noise ratio (SNR) higher than 25 dB [36]. The BER performance is determined by the communication environment and SNR, therefore the SNR over distance and the BER under different SNRs should be measured. Figure 10a shows the measured SNR under different distances; it can be seen that within the communication distance of 10.3 m, the measured SNR is higher than 26 dB. Figure 10b shows the measured BER under different SNR. As seen in the figure, when the SNR is higher than 25 dB, the measured BER is less than 0.001. Therefore, the proposed wireless sensor works well in the test scenario.
Figure 10. 10. (a) (a) Measured Measured signal signal to Figure to noise noise ratio ratio (SNR) (SNR) under under different different distances; distances; (b) (b)measured measuredbit biterror error rate (BER) under different SNRs. rate (BER) under different SNRs.
In order to verify the estimated performances of the proposed method, we tested the wireless In order to verify the estimated performances of the proposed method, we tested the wireless sensor at the underground mine site. Meanwhile, a commercial gas concentration detector was put sensor at the underground mine site. Meanwhile, a commercial gas concentration detector was put in the same spot as a reference. The proposed approach was employed to predict the methane gas in the same spot as a reference. The proposed approach was employed to predict the methane gas density, and the measurement results are shown in Figure 11. As can be seen, the predicted results density, and the measurement results are shown in Figure 11. As can be seen, the predicted results are are consistent with the measured results and the maximum error is 0.42 ppm within 100 min. consistent with the measured results and the maximum error is 0.42 ppm within 100 min. In general, the available historical data is also one of the factors that affect the prediction accuracy. Therefore, we measured the prediction error under different available historical data and the measured results are shown in Figure 12. At the beginning of prediction, the LS-SVR model is trained by little historical data, which leads to big prediction error. With the increase of available historical data, the error is quickly decreased to 0.00012. Thus, it can be deduced that the proposed method can achieve high prediction accuracy with sufficient historical data.
Figure 11. Comparison of measured results and predicted results.
In general, the available historical data is also one of the factors that affect the prediction accuracy. Therefore, we measured the prediction error under different available historical data and
In order to verify the estimated performances of the proposed method, we tested the wireless sensor at the underground mine site. Meanwhile, a commercial gas concentration detector was put in the same spot as a reference. The proposed approach was employed to predict the methane gas density, and the measurement results are shown in Figure 11. As can be seen, the predicted results Electronics 2018, 7, 69 10 of 13 are consistent with the measured results and the maximum error is 0.42 ppm within 100 min.
Figure 11. Comparison of measured measured results results and and predicted predicted results. results. Figure 11. Electronics 2018, 7, x FOR PEER REVIEW
10 of 12
In general, the available historical data is also one of the factors that affect the prediction accuracy. Therefore, we measured the prediction error under different available historical data and the measured results are shown in Figure 12. At the beginning of prediction, the LS-SVR model is trained by little historical data, which leads to big prediction error. With the increase of available historical data, the error is quickly decreased to 0.00012. Thus, it can be deduced that the proposed method can achieve high prediction accuracy with sufficient historical data. To better verify the priorities of the proposed method, its performance is compared with several wireless sensors which are introduced for methane gas density measurement. Table 2 shows the comparison results. As can be seen, the exploited wireless sensor shows the advantages including low cost, additional ID information, additional predicting ability, longer communication distance, and higher accuracy compared with the other reported wireless sensors.
Figure (LS-SVR). Figure 12. 12. Prediction Prediction error error of of least least squares squares support support vector vector regression regression (LS-SVR). Table 2. Performances comparison of different wireless sensors.
To better verify the priorities of the proposed method, its performance is compared with several Reported sensors Method which Cost are introduced Memory Function Accuracy wireless for methane gas density measurement.Range Table(m)2 shows the [8] high gasexploited density monitoring / high comparison results. As can be methane seen, the wireless sensor shows the advantages including [9] high methane gas density monitoring / medium low cost, additional ID information, additional predicting ability,monitoring longer communication distance, and [15] high methane gas density / high higher accuracy compared with the other reported wireless sensors. This work low methane gas density and ID message monitoring and predicting 17 high 5. Conclusions
Table 2. Performances comparison of different wireless sensors.
Reported Method the Costadvantages ofMemory Function Range (m) Accuracy Considering integrating RFID and WSN, a novel monitoring and predicting high methane gasdensity density is introducedmonitoring / high method [8] for underground methane gas in this paper. The proposed wireless [9] high methane gas density monitoring / medium sensor is[15] based on high passive RFID technology and consists of amonitoring communication section, RF front-end methane gas density / high section, section. The data transmitted by the wireless contains miscellaneous noises, Thisand workdigital low methane gas density and ID message monitoringsensor and predicting 17 high which makes it difficult to realize accurate monitoring and prediction. Hence, a deep learning approach is employ to extract robust features from the raw data. The experimental results 5. Conclusions demonstrate that the introduced wireless sensor achieves a maximum communication distance of Considering the advantages of integrating RFID and WSN, a novel monitoring and predicting 10.3 m and can accurately measure and predict the underground methane gas density. method for underground methane gas density is introduced in this paper. The proposed wireless sensor Contributions: is based on passive RFID and of a communication section, RF front-end Author C.Z. and Y.F. technology are responsible forconsists the algorithm design. F.D. and X.W. are responsible for
the design of the RFID sensor tag. All authors provided help in the revision of this manuscript. Acknowledgments: This work was supported by Natural Science Foundation of China (51767006, 51464015), Natural Science Foundation of Jiangxi Province (20171BAB206045), and Science and Technology Project of Education Department of Jiangxi Province (GJJ160491, GJJ170378).
Electronics 2018, 7, 69
11 of 13
section, and digital section. The data transmitted by the wireless sensor contains miscellaneous noises, which makes it difficult to realize accurate monitoring and prediction. Hence, a deep learning approach is employ to extract robust features from the raw data. The experimental results demonstrate that the introduced wireless sensor achieves a maximum communication distance of 10.3 m and can accurately measure and predict the underground methane gas density. Author Contributions: C.Z. and Y.F. are responsible for the algorithm design. F.D. and X.W. are responsible for the design of the RFID sensor tag. All authors provided help in the revision of this manuscript. Acknowledgments: This work was supported by Natural Science Foundation of China (51767006, 51464015), Natural Science Foundation of Jiangxi Province (20171BAB206045), and Science and Technology Project of Education Department of Jiangxi Province (GJJ160491, GJJ170378). Conflicts of Interest: The authors declare no conflict of interest.
References 1.
2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.
14.
15.
16.
Zhang, Y.; Yang, W.; Han, D.; Kim, Y.-I. An integrated environment monitoring system for underground coal mines—Wireless Sensor Network subsystem with multi-parameter monitoring. Sensors 2014, 14, 13149–13170. [CrossRef] [PubMed] Ni, L.M.; Liu, Y.; Zhu, Y. China’s National Research Project on Wireless Sensor Networks. IEEE Wirel. Commun. 2011, 14, 78–83. [CrossRef] Li, M.; Liu, Y. Underground Coal Mine Monitoring with Wireless Sensor Networks. ACM Trans. Sens. Netw. 2012, 5, 1–29. [CrossRef] Gao, Y.; Fu, G.; Nieto, A. A comparative study of gas explosion occurrences and causes in China and the United States. Int. J. Min. Reclam. Environ. 2015, 30, 269–278. [CrossRef] Wang, H.; Cheng, Y.; Yuan, L. Gas outburst disasters and the mining technology of key protective seam in coal seam group in the huainan coalfield. Nat. Hazards 2013, 67, 763. [CrossRef] Joy, G.J. Evaluation of the approach to respirable quartz exposure control in us coal mines. J. Occup. Environ. Hyg. 2012, 9, 65–68. [CrossRef] [PubMed] CDC. Center for Disease Control and Prevention Mining Facilities. Available online: http://www.cdc.gov/ niosh/mining/statistics/allmining.html (accessed on 6 September 2017). Jo, B.W.; Khan, R.M.A. An Event Reporting and Early-Warning Safety System Based on the Internet of Things for Underground Coal Mines: A. Case Study. Appl. Sci. 2017, 7, 925. Osunmakinde, I.O. Towards safety from toxic gases in underground mines using wireless sensor networks and ambient intelligence. Int. J. Distrib. Sens. Netw. 2013, 9. [CrossRef] Zhao, J.; Gao, J.; Zhao, F.; Liu, Y. A Search-and-Rescue Robot System for Remotely Sensing the Underground Coal Mine Environment. Sensors 2017, 17, 2426. [CrossRef] [PubMed] Byun, J.; Jeon, B.; Noh, J.; Kim, Y.; Park, S. An intelligent self-adjusting sensor for smart home services based on ZigBee communications. IEEE Trans. Consum. Electron. 2012, 58, 794–802. [CrossRef] Kumar, A.; Hancke, G.P. Energy efficient environment monitoring system based on the IEEE 802.15.4 standard for low cost requirements. IEEE Sens. J. 2014, 14, 2557–2566. [CrossRef] Kovaˇci´c, M.; Vraži´c, M.; Gasparac, I. Bluetooth wireless communication and 1-wire digital temperature sensors in synchronous machine rotor temperature measurement. In Proceedings of the 14th International Power Electronics Motion Control Conference, Ohrid, Makedonija, 6–8 September 2010; pp. T7-25–T7-28. Chehri, A.; Farjow, W.; Mouftah, H.T.; Fernando, X. Design of wireless sensor network for mine safety monitoring. In Proceedings of the 2011 24th Canadian Conference on Electrical and Computer Engineering, Niagara Falls, ON, Canada, 8–11 May 2011. Moridi, M.A.; Kawamura, Y.; Sharifzadeh, M.; Chanda, E.K.; Jang, H. An investigation of underground monitoring and communication system based on radio waves attenuation using ZigBee. Tunn. Undergr. Space Technol. 2014, 43, 362–369. [CrossRef] Bhattacharjee, S.; Roy, P.; Ghosh, S.; Misra, S.; Obaidat, M.S. Wireless sensor network-based fire detection, alarming, monitoring and prevention system for Bord-and-Pillar coal mines. J. Syst. Softw. 2012, 85, 571–581. [CrossRef]
Electronics 2018, 7, 69
17. 18. 19. 20. 21.
22.
23. 24. 25. 26. 27. 28. 29. 30. 31. 32.
33. 34. 35. 36.
37.
38.
39.
12 of 13
Deng, F.; He, Y.; Li, B.; Song, Y.; Wu, X. Design of a Slotted Chipless RFID Humidity Sensor Tag. Sens. Act. B Chem. 2018, 264, 255–262. [CrossRef] Deng, F.; He, Y.; Li, B.; Zuo, L.; Wu, X.; Fu, Z. A CMOS pressure sensor tag chip for passive wireless applications. Sensors 2015, 15, 6872–6884. [CrossRef] [PubMed] Deng, F.; He, Y.; Li, B.; Zhang, L.; Wu, X.; Fu, Z.; Zuo, L. Design of an Embedded CMOS Temperature Sensor for Passive RFID Tag Chips. Sensors 2015, 15, 11442–11453. [CrossRef] [PubMed] Catarinucci, L.; Donno, D.D.; Mainetti, L.; Palano, L.; Patrono, L.; Stefanizzi, M.L.; Tarricone, L. An IoT-Aware Architecture for Smart Healthcare Systems. IEEE Int. Things J. 2015, 2, 515–526. [CrossRef] Mainetti, L.; Palano, L.; Patrono, L.; Vergallo, R.; Stefanizzi, M.L. Integration of RFID and WSN technologies in a Smart Parking System. In Proceedings of the 2014 22nd International Conference on Software, Telecommunications and Computer Networks, Split, Croatia, 17–19 September 2014; pp. 104–110. Mirshahi, S.; Uysal, S.; Akbari, A. Integration of RFID and WSN for supply chain intelligence system. In Proceedings of the 2013 International Conference on Electronics, Computers and Artificial Intelligence, Pitesti, Romania, 27–29 June 2013; pp. 1–6. Amin, E.M.; Saha, J.K.; Karmakar, N.C. Smart Sensing Materials for Low-Cost Chipless RFID Sensor. IEEE Sens. J. 2014, 14, 2198–2207. [CrossRef] Donno, D.D.; Catarinucci, L.; Tarricone, L. RAMSES: RFID Augmented Module for Smart Environmental Sensing. IEEE Trans. Instrum. Meas. 2014, 63, 1701–1708. [CrossRef] Zhou, S.; Sheng, W.; Deng, F.; Wu, X.; Fu, Z. A Novel Passive Wireless Sensing Method for Concrete Chloride Ion Concentration Monitoring. Sensors 2017, 17, 2871. [CrossRef] [PubMed] Wang, J.; He, Q. Wavelet packet envelope manifold for fault diagnosis of rolling element bearings. IEEE Trans. Instrum. Meas. 2016, 65, 2515–2526. [CrossRef] He, Q.; Song, H.; Ding, X. Sparse signal reconstruction based on time–frequency manifold for rolling element bearing fault signature enhancement. IEEE Trans. Instrum. Meas. 2016, 65, 482–491. [CrossRef] Ren, L.; Lv, W.; Jiang, S.; Xiao, Y. Fault diagnosis using a joint model based on sparse representation and SVM. IEEE Trans. Instrum. Meas. 2016, 65, 2313–2320. [CrossRef] Wang, Y.; Wang, X.; Liu, W. Unsupervised local deep feature for image recognition. Inf. Sci. 2016, 351, 67–75. [CrossRef] Zhan, S.; Tao, Q.-Q.; Li, X.-H. Face detection using representation learning. Neurocomputing 2016, 187, 19–26. [CrossRef] Siniscalchi, S.M.; Yu, D.; Deng, L.; Lee, C.-H. Exploiting deep neural networks for detection-based speech recognition. Neurocomputing 2013, 106, 148–157. [CrossRef] Jiang, G.; He, H.; Xie, P.; Tang, Y. Stacked Multilevel-Denoising Autoencoders: A New Representation Learning Approach for Wind Turbine Gearbox Fault Diagnosis. IEEE Trans. Instrum. Meas. 2017, 99, 1–12. [CrossRef] Hu, Y.; Zhao, H.F.; Wang, Z.G. Sign Language Fingerspelling Recognition Using Depth Information and Deep Belief Networks. Int. J. Pattern Recognit. Artif. Intell. 2018, 32, 1850018. [CrossRef] Jing, L.; Zhao, M.; Li, P.; Xu, X. A convolutional neural network based feature learning and fault diagnosis method for the condition monitoring of gearbox. Measurement 2017, 111, 1–10. [CrossRef] Cha, Y.J.; Choi, W.; Büyüköztürk, O. Deep Learning-Based Crack Damage Detection Using Convolutional Neural Networks. Comput. Aided Civil Infrastr. Eng. 2017, 32, 361–378. [CrossRef] EPCglobal Inc. EPC Radio-Frequency Identity Protocols Generation-2 UHF RFID. 2008. Available online: https://www.gs1.org/sites/default/files/docs/epc/Gen2_Protocol_Standard.pdf (accessed on 12 May 2018). Liu, B.H.; Nguyen, N.T.; Pham, V.T.; Yeh, Y.H. A Maximum-Weight-Independent-Set-Based Algorithm for Reader-Coverage Collision Avoidance Arrangement in RFID Networks. IEEE Sens. J. 2016, 16, 1342–1350. [CrossRef] Nguyen, N.T.; Liu, B.H.; Pham, V.T. A dynamic-range-based algorithm for reader-tag collision avoidance deployment in RFID networks. In Proceedings of the 2016 International Conference on Electronics, Information, and Communications, Da Nang, Vietnam, 27–30 January 2016; pp. 1–4. Gu, B.; Sheng, V.S.; Tay, K.Y.; Romano, W.; Li, S. Incremental Support Vector Learning for Ordinal Regression. IEEE Trans. Neural Netw. Learn. Syst. 2015, 26, 1403. [CrossRef] [PubMed]
Electronics 2018, 7, 69
40. 41.
13 of 13
Gu, B.; Sheng, V.S. A Robust Regularization Path Algorithm for ν-Support Vector Classification. IEEE Trans. Neural Netw. Learn. Syst. 2016, 99, 1–8. [CrossRef] [PubMed] Wang, T.; He, Y.; Luo, Q.; Deng, F.; Zhang, C. Self-Powered RFID Sensor Tag for Fault Diagnosis and Prognosis of Transformer Winding. IEEE Sens. J. 2017, 17, 6418–6430. [CrossRef] © 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
