CN115128702A - Composite microwave sensor and detection method - Google Patents

Abstract

The invention discloses a composite microwave sensor and a detection method, which belong to the technical field of microwave detection. The method analyzes the offset change of the microwave resonance peak data of humidity, rainfall and mass fog relative to the drying cavity, and at the same time conducts cross-combination of the interaction between the three detection variable data of humidity, rainfall and mass fog. By analyzing and establishing the corresponding back-propagation neural network, more accurate humidity and rainfall values can be obtained and more accurate judgment of the fog size can be made. This application adopts a 5th-order back-propagation network multiple regression model. Compared with the ordinary back-propagation neural network, this model defines the interaction between humidity, rainfall, and fog as the demand function in the last layer, and the parameter The relationship between them is reflected in the parameter weight of the demand function, and has the interconnectivity of multiple parameters, so as to achieve a more accurate prediction of humidity, rainfall, and fog.

Description

A composite microwave sensor and detection method

technical field

The invention relates to a composite microwave sensor and a detection method, belonging to the technical field of microwave detection.

Background technique

Humidity sensor is a detection device that converts environmental humidity information into electrical signals based on sensitive components. It is widely used in humidity monitoring in industrial production, agricultural planting, medical monitoring, food storage and other links. Traditional humidity sensors can be divided into resistive, capacitive, optical, surface acoustic wave and other types. However, most of the existing microwave-based humidity sensors only detect a single parameter to characterize the humidity, which has the problem of poor stability, and the detection accuracy needs to be further improved.

Fog is highly regional and difficult to predict and forecast, especially on expressways, fog can lead to sudden changes in visibility, which is extremely harmful to expressway traffic safety and can easily lead to major traffic accidents. The coverage area of the fog is also inconsistent. Generally speaking, the coverage area of a large fog is about five kilometers long, and the small fog is only one kilometer. In recent years, more in-depth studies have been conducted on the characteristics of fog, the formation principle of fog, and grade evaluation. According to the characteristic parameters of fog, there are now reference indicators for classifying fog levels, which further expands the level of fog. Classification. Most of the existing fog alarm devices have the characteristics of complex structure and poor portability, and are difficult to integrate.

The rainfall judgment measuring instrument is a tool used to measure the size of the local rainfall, and calculate the specific rainfall at a certain time in the local area accordingly. Predicting the rainfall and its changes are very important in the field of hydrological cycle. However, the existing rainfall judgment measuring instruments have disadvantages such as large and complex systems, low operability, and poor portability. Therefore, how to improve the operability and portability of the rainfall judgment measuring instrument is an urgent problem to be solved.

Therefore, how to efficiently and accurately judge the environmental humidity, fog concentration and rainfall, and at the same time detect the existence of fog and give early warning to highway vehicles, is a topic worthy of in-depth study. Therefore, a three-in-one sensor that can detect both humidity and fog, and judge rainfall is needed.

SUMMARY OF THE INVENTION

In order to realize the detection of humidity, rainfall and fog at the same time, the present invention provides a composite microwave sensor and a detection method. By providing a microwave sensor device with four resonance units, the detection of humidity, rainfall and fog can be realized simultaneously. Detection, and when performing detection, cross-joint analysis is performed on the interaction between the three detection variable data of humidity, rainfall, and fog, and a corresponding back-propagation neural network is established. Compared with the general back-propagation Neural network, the back-propagation neural network established in this application defines the interaction between humidity, rainfall, and fog as a demand function in the last layer, and reflects the interaction between parameters on the parameter weight of the demand function , so it has multi-parameter interconnectivity, and then more accurate humidity and rainfall values can be obtained, and a more accurate judgment of the size of the fog cloud can be made.

A composite microwave sensor, the composite microwave sensor is used to simultaneously detect humidity, rainfall and fog in the environment to be detected; the composite microwave sensor includes a feeder and four sensing units; each sensor The unit consists of two split resonant rings; among them, the sensing unit 1 is a cavity comparison unit, the sensing unit 2 is a humidity sensing unit, the sensing unit 3 is a mass fog detection unit, and the sensing unit 4 is a rainfall detection unit, The four sensing units correspond to different resonance peaks according to different sizes.

Optionally, each sensing unit in the composite microwave sensor generates resonance peak shifts of different magnitudes for changes in humidity and water vapor saturation in the environment to be detected.

A detection method based on the above-mentioned composite microwave sensor, the method comprising:

The composite microwave sensor is placed in the environment to be measured, the position of the resonance peak of each sensing unit is obtained, and the frequency information corresponding to the peak value is converted into a voltage signal;

Input the converted voltage signal into the trained back-propagation neural network to obtain the humidity value and water vapor saturation value of the environment to be measured, and determine the occurrence of rainfall and fog according to the humidity value and water vapor saturation value.

Optionally, the back-propagation neural network is a fifth-order back-propagation network, and the last layer of the network adopts a multiple regression model, and the demand function of the multiple regression model is defined as:

y=θ ₁ x ₁ +θ ₂ x ₂ +θ ₃ x ₃ +θ ₄ x ₄ +θ ₅ (6)

表示训练样本对应的湿度值和水汽饱和度值，则有：where θ ₅ represents the deviation value in the multiple regression model; where θ=[θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ ] is the cross-correlation coefficient required for training, which is used to represent the resonance of the four sensing units The relationship between the peak change and the final predicted value is defined as X=[x ₁ , x ₂ , x ₃ , x ₄ ] for all training sample inputs, which are the corresponding peak values of the resonance peaks of the four sensing units, respectively. The voltage signal obtained by converting the frequency information, the label of the last layer of training output is defined as

Indicates the humidity value and water vapor saturation value corresponding to the training sample, there are:

Among them, the subscript i represents the label of the training sample;

The final calculation to obtain the optimal parameter matrix θ is,

θ=(X ^T X) ^-1 X ^T y (8)

where T stands for transpose.

Optionally, in the training process of the back-propagation neural network, the loss function is:

Where n is the total number of training samples X, y=y(x) is the desired output, L is the number of layers of the back-propagation neural network, and a ^L (x) is the output vector of the back-propagation neural network;

的值越来越小，每一层神经网络的输出层误差为：error during each training

The value of is getting smaller and smaller, and the output layer error of each layer of neural network is:

Equation (2) is the matrix form of the output error, σ'(z ^l ) is the partial derivative of a neuron activation function to the l layer, l={1,2,...,L};

The error transfer equation between layers is:

δ ^l =((w ^l+1 ) ^T δ ^l+1 )⊙σ′(z ^l ) (3)

Combine formula (2) and formula (3) to calculate the error of any layer in the neural network, that is, first calculate the l layer, and then decrease layer by layer until the first layer is calculated;

The parameters of each layer of the back-propagation neural network include the weight w and the bias b, and the change rate of the cost function to the weight w is,

The rate of change of the cost function to the bias b is,

和

后，使用梯度下降法对参数进行一轮的更新，直至代价函数对参数的偏导数不断变小，最终确定反向传播神经网络的参数权重W和偏置B，得到训练好的反向传播神经网络。get

and

Then, use the gradient descent method to update the parameters for one round until the partial derivative of the cost function to the parameters keeps getting smaller, and finally determine the parameter weight W and bias B of the back-propagation neural network, and get the trained back-propagation neural network. network.

Optionally, the acquiring the resonance peak position of each sensing unit includes:

Perform differential calculation for each frequency point, and use the feature that the difference value before the resonance peak is greater than 0 and the difference after the resonance peak is less than 0 to determine the position of the resonance peak, so as to obtain the frequency value at the resonance peak.

The beneficial effects of the present invention are:

Different from the traditional microwave humidity sensor, the present invention proposes a microwave sensing system integrating humidity, rainfall and mass fog detection, which has a simplified structure, a high degree of integration and high measurement accuracy. The four resonance units in the microwave sensor device correspond to four different measurement variables respectively, and the corresponding environmental conditions can be judged and calculated through multiple parameters. The four detection variables correspond to humidity, rainfall, fog and dry cavity respectively. The interaction between the three detection variable data is cross-joint analysis and the corresponding back-propagation neural network is established, so that more accurate humidity and rainfall values can be obtained, and more accurate fog cloud size can be obtained. judge. Compared with the existing traditional humidity and rainfall sensors, the present invention is more accurate and reliable in terms of obtaining the results of humidity and rainfall by simply detecting a single parameter variable. The reliability of the results can be greatly improved, and the obtained result interval can further predict the future humidity and rainfall. Compared with the existing visual fog fog detection system, the detection of fog fog in the present invention not only compares and analyzes with the dry environment, but also combines the environmental humidity and rainfall, so that the analysis result of the fog size is more accurate than visual detection. In order to be accurate, and to be able to measure the concentration of the fog.

When processing the obtained humidity, rainfall and mass fog data, the present invention adopts the 5th-order back propagation network multiple regression model. Compared with the ordinary back propagation neural network, this algorithm model combines humidity, rainfall, The interaction between the fogs is defined as the demand function, and the interaction between the parameters is reflected in the parameter weight of the demand function, and the model is finally obtained through parameter training. Compared with the traditional back-propagation neural network algorithm, the algorithm model of the invention has more multi-parameter interconnectivity, and at the same time, for complex and changeable environmental conditions, the results obtained by the model are more reliable and the accuracy is greatly improved.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a structural diagram of a sensing unit of a composite microwave sensor provided in an embodiment of the application;

FIG. 2 is a schematic diagram of S-parameters of each sensing unit of a composite microwave sensor provided in an embodiment of the application;

FIG. 3 is an equivalent circuit diagram of a sensing unit of a composite microwave sensor provided in an embodiment of the application;

FIG. 4 is an equivalent circuit diagram of a composite microwave sensor provided in an embodiment of the present application.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

Example 1:

This embodiment provides a composite microwave sensor, see FIG. 1 , the composite microwave sensor includes a feeder and four sensing units; each sensing unit is composed of two split resonant rings; wherein, the sensing unit 1 is empty Cavity control unit, sensing unit 2 is a humidity sensing unit, sensing unit 3 is a fog detection unit, sensing unit 4 is a rainfall detection unit, and the four sensing units correspond to different resonance peaks.

Each sensing unit in the composite microwave sensor generates frequency offsets of different sizes according to the changes of humidity and water vapor saturation in the environment to be detected.

In the preparation process of the composite microwave sensor, wet etching technology can be used to etch the pattern of the sensing unit on a Teflon substrate with a thickness of 2.54 mm, as shown in FIG. 1 . It includes a feeder and four sensing units, the four sensing units are 1, 2, 3, and 4 respectively, and each sensing unit is composed of two split resonant rings.

In the subsequent simulation of the composite microwave sensor in this embodiment, the four sensing units resonate at frequencies of 3.72 GHz, 4.02 GHz, 4.37 GHz, and 4.78 GHz, respectively, as shown in FIG. 2 . The sensing unit 1 is a cavity control unit, which is used as a control for humidity, fog, and rainfall detection, and the resonance peak frequency generated by it is the lowest, which corresponds to the resonance sensitive peak 5; The peak corresponds to the resonance sensitive peak 6; the sensing unit 3 is a fog detection unit, and its resonance peak corresponds to the resonance sensitive peak 7; the sensing unit 4 is a rainfall detection resonance peak, which is specifically detected by the water vapor saturation in the environment. , and its resonance peak corresponds to the resonance sensitive peak 8 . The humidity value and water vapor saturation value in the environment to be detected are determined through the resonance peak offset detected by the four sensing units, and then the rainfall and fog conditions are determined.

FIG. 3 is an equivalent circuit diagram of four sensing units. The sensing unit is a passive device, so it is equivalent to the resonance of capacitance and inductance.

In the microwave sensor control circuit of Figure 4, the power supply circuit is supplied with 220V AC voltage by VSS, which is regulated by the voltage drop capacitor C1 and diode D2, and then rectified by D1 and filtered by C2 and C3, and then becomes the source of the subsequent sensor detection module. Applicable supply voltage. IC1 is a microwave sensor processing module, which uses the differential frequency sweep algorithm to traverse the cyclic valley calculation of the S-parameter map detected by the sensor, calculate the position of each resonance peak, and convert the frequency information corresponding to the peak into a voltage signal. Output; IC2 is an ADC conversion module, which receives three voltage signals and converts them into digital signals and sends them to the IC3 module for analysis; IC3 is an FPGA (Zynq-7000) processing module, which uses the corresponding embedded algorithm to get the result. The digital signal is used for modular calculation and finally obtains the environmental humidity, rainfall and fog information.

Specifically, when the humidity, water vapor saturation, and fog in the environment change, it will cause corresponding changes in each sensing unit, that is, the resonance peak corresponding to each sensing unit will produce corresponding changes on the sensor S-parameter diagram. displacement change. The microwave sensor processing module uses the differential frequency sweep algorithm to perform differential calculation on each frequency point from 3.5MHZ on the S-parameter diagram. The difference value before the resonance point is greater than 0, and the difference after the resonance point is less than 0 to determine the resonance point. The position of the resonance point, and the auxiliary detection method of windowing calculation is used for the resonance point. A frequency window is added on both sides of the peak value, and the peak with a large absolute value of the difference in the frequency window is the required resonance peak.

When there is no major change in the environment, the sensing unit transmits the stably generated signal to the microwave sensing processing module IC1, so that Q1 generates a base-level high level and then turns on, and pins 1, 2, and 3 become low levels. , the ADC conversion module IC2 detects the corresponding analog voltage signal and outputs the corresponding digital signal at pins 4, 5, and 6. After limiting the current, it is used as the three-port input of the FPGA (Zynq-7000) processing module, and it is displayed as no at the terminal. Water vapor, fog and rain occur.

When the humidity in the environment changes greatly and is accompanied by rainfall, the resonant peaks of the humidity sensing unit and the rainfall detection unit have different frequency shifts respectively. The resonance point is detected and converted into a correspondingly changing voltage signal, and then the response signal is output to make the transistor Q1 completely cut off, so that the pins 1 and 2 become high level. At this time, the ADC conversion module IC2 converts the input analog voltage signal into Digital signal, in the next clock cycle, pins 4 and 5 output the corresponding digital level, the input analog voltage value is related to the change of the resonance peak monitored by the sensor detection module, that is, when the ambient humidity becomes larger , the voltage value increases accordingly. This changed voltage value is converted into a correspondingly changed digital signal by the ADC conversion module IC2, and is input to the FPGA (Zynq-7000) processing module IC3. -7000), through relevant algorithm analysis, it is finally displayed at the terminal that the environment is in a state of high water vapor saturation and rainfall occurs, and a rainfall alarm is performed.

When a large fog is generated in the environment, because the appearance of fog is often accompanied by the increase of water vapor saturation and the increase of environmental humidity, the resonance peak of the fog sensor unit will be greatly shifted, and the rainfall sensor unit and the Smaller shift in the resonant peak of the humidity sensing unit. Correspondingly, the microwave sensor processing module IC1 detects the three changed resonance peaks through the differential frequency sweep algorithm and converts them into voltage signals to make pins 1, 2, and 3 become high levels. The ADC digital-to-analog conversion module IC2 converts these peaks. A changing analog voltage is converted to a changing digital level and output at pins 4, 5, and 6 after the next clock cycle. The three-way voltage enters the FPGA (Zynq-7000) processing module IC3 after current limiting and filtering, and displays the fog on the terminal through the relevant algorithm, and the weak change of the input high and low level signals will affect the analysis results of the IC3 module and The reaction is in a cloudy fog condition.

Embodiment 2:

This embodiment provides a detection method for a composite microwave sensor. The method is based on the composite microwave sensor described in Embodiment 1 to realize the detection of environmental humidity, rainfall, and fog in the environment to be measured. The method includes:

The composite microwave sensor is placed in the environment to be measured, the position of the resonance peak of each sensing unit is obtained, and the frequency information corresponding to the peak value is converted into a voltage signal;

Input the converted voltage signal into the trained back-propagation neural network to obtain the humidity value and water vapor saturation value of the environment to be measured, and determine the occurrence of rainfall and fog according to the humidity value and water vapor saturation value.

In this application, a 5-layer back-propagation neural network is established for the composite microwave sensor with 4 sensing units. For the three-input voltage signal values of humidity, water vapor saturation, and fog conditions, a three-output value of ambient humidity, whether it rains, and fog conditions are given and mapped with the input, and the coefficients of this mapping are the pre-defined values of this application. The network parameter W that needs to be trained, after the mapped network parameters are determined, the corresponding input voltage value can be calculated to obtain the ambient humidity, rainfall and fog of the environment to be measured. At the same time, the established multivariate regression model is added to the training network to analyze and process the interaction between humidity, rainfall and fog, so that the final data reflects the correlation between the three variables measured, thereby greatly improving the accuracy of the results. degree and reliability.

This application adopts the back-propagation algorithm to train the network parameter W, and first defines a loss function, as follows,

的值越来越小，即可以得到每一层神经网络的输出层误差为，Where n is the total number of training samples x, y=y(x) is the desired output, L is the number of layers of the network a ^L (x) is the output vector of the network. In order to calculate the specific value of the network parameters, we need to make the error in the process of each training

The value of is getting smaller and smaller, that is, the output layer error of each layer of neural network can be obtained as,

The above formula is the matrix form of the output error, and σ'(z ^l ) is the partial derivative of a neuron activation function to the l layer. At the same time, the error transfer equation between layers can also be obtained:

δ ^l =((w ^l+1 ) ^T δ ^l+1 )⊙σ′(z ^l ) (3)

This equation shows that the error δ ^{l of the lth layer can be calculated by the error δl+1 of} the l+1th layer, and the error of any layer in the neural network can be calculated by combining the formula (2) and the formula (3). That is, the δ ^l layer is calculated first, and then it is decreased layer by layer until the first layer is calculated.

The parameters of each layer of the back-propagation neural network include the weight w and the bias b, and the change rate of the cost function to the weight w is,

The rate of change of the cost function to the bias b is,

和

后，使用梯度下降法对参数进行一轮的更新，直至代价函数对参数的偏导数不断变小，即模型不断收敛，最终可得到所需的参数网络。when getting

and

Then, use the gradient descent method to update the parameters for one round until the partial derivative of the cost function to the parameters continues to become smaller, that is, the model continues to converge, and finally the required parameter network can be obtained.

After determining the parameter weight w and bias b of each layer, the corresponding matrix is the parameter weight W and bias B of the back-propagation neural network.

Different from the traditional back-propagation neural network, in the present invention, the multi-regression model is used for the last layer of the back-propagation model, and the demand function of the multi-regression model is defined as:

y=θ ₁ x ₁ +θ ₂ x ₂ +θ ₃ x ₃ +θ ₄ x ₄ +θ ₅ (6)

表示训练样本对应的湿度值和水汽饱和度值，则有，where θ ₅ represents the bias value in the multiple regression model, i.e. the bias b of the 5th layer network. Where θ=[θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ ] is the cross-correlation coefficient required for training, and all training sample inputs are defined as X=[x ₁ , x ₂ , x ₃ , x ₄ ], the label of the last layer training output is defined as

represents the humidity value and water vapor saturation value corresponding to the training sample, then there are,

Among them, the subscript i represents the label of the training input value, that is, the i-th group of training samples; the optimal parameter matrix θ is finally calculated as:

θ=(X ^T X) ^-1 X ^T y (8)

When the training of the neural network algorithm is completed, a set of weight and bias parameter matrices of the neural network are obtained. In the subsequent use process, the algorithm operation can be completed by receiving the digital signals of the three inputs. The humidity, precipitation and fog of the environment can be obtained at the terminal. This result not only uses the dry cavity as a comparison, but also integrates humidity, rainfall, and fog. The interaction between them can achieve very high accuracy.

The above is an example of system detection for a single detection cycle, but in practical applications, the humidity and water vapor saturation of the environment change in real time, that is, the entire detection system detects and updates the environmental conditions in real time, with extremely high real-time performance.

Some steps in the embodiments of the present invention may be implemented by software, and corresponding software programs may be stored in a readable storage medium, such as an optical disc or a hard disk.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within the range.

Claims (6)

1. A composite microwave sensor, characterized in that the composite microwave sensor is used to simultaneously detect humidity, rainfall and fog in the environment to be detected; the composite microwave sensor comprises a feeder and four sensors. Each sensing unit consists of two split resonant rings; among them, sensing unit 1 is a cavity control unit, sensing unit 2 is a humidity sensing unit, sensing unit 3 is a fog detection unit, and sensing unit 4 is a rainfall detection unit, and the four sensing units correspond to different resonance peaks according to different sizes. 2 . The composite microwave sensor according to claim 1 , wherein each sensing unit in the composite microwave sensor generates resonance peaks of different sizes according to changes in humidity and water vapor saturation in the environment to be detected. 3 . offset. 3. A detection method based on the composite microwave sensor according to claim 1 or 2, wherein the method comprises: The composite microwave sensor is placed in the environment to be measured, the position of the resonance peak of each sensing unit is obtained, and the frequency information corresponding to the peak is converted into a voltage signal; Input the converted voltage signal into the trained back-propagation neural network to obtain the humidity value and water vapor saturation value of the environment to be measured, and determine the occurrence of rainfall and fog according to the humidity value and water vapor saturation value. 4. method according to claim 3, is characterized in that, described back-propagation neural network is 5th-order back-propagation network, and last layer network adopts multiple regression model, and the demand function that defines multiple regression model is: y=θ ₁ x ₁ +θ ₂ x ₂ +θ ₃ x ₃ +θ ₄ x ₄ +θ ₅ (6) where θ ₅ represents the deviation value in the multiple regression model; where θ=[θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ ] is the cross-correlation coefficient required for training, which is used to represent the resonance of the four sensing units The relationship between the peak change and the final predicted value is defined as X=[x ₁ , x ₂ , x ₃ , x ₄ ] for all training sample inputs, which are the corresponding peak values of the resonance peaks of the four sensing units, respectively. The voltage signal obtained by converting the frequency information, the label of the last layer of training output is defined as

Indicates the humidity value and water vapor saturation value corresponding to the training sample, there are:

Among them, the subscript i represents the label of the training sample; The final calculation to obtain the optimal parameter matrix θ is, θ=(X ^T X) ^-1 X ^T y (8) where T stands for transpose. 5. The method according to claim 4, wherein, in the training process of the back-propagation neural network, the loss function is:

Where n is the total number of training samples X, y=y(x) is the desired output, L is the number of layers of the back-propagation neural network, and a ^L (x) is the output vector of the back-propagation neural network; error during each training

The value of is getting smaller and smaller, and the output layer error of each layer of neural network is:

Equation (2) is the matrix form of the output error, σ'(z ^l ) is the partial derivative of a neuron activation function to the l layer, l={1,2,...,L}; The error transfer equation between layers is: δ ^l =((w ^l+1 ) ^T δ ^l+1 )⊙σ′(z ^l ) (3) Combine formula (2) and formula (3) to calculate the error of any layer in the neural network, that is, first calculate the l layer, and then decrease layer by layer until the first layer is calculated; The parameters of each layer of the back-propagation neural network include the weight w and the bias b, and the change rate of the cost function to the weight w is,

The rate of change of the cost function to the bias b is,

get

and

Then, use the gradient descent method to update the parameters for one round until the partial derivative of the cost function to the parameters keeps getting smaller, and finally determine the parameter weight W and bias B of the back-propagation neural network, and get the trained back-propagation neural network. network. 6. The method according to claim 5, wherein the acquiring the resonance peak position of each sensing unit comprises: Perform differential calculation for each frequency point, and use the feature that the difference value before the resonance peak is greater than 0 and the difference after the resonance peak is less than 0 to determine the position of the resonance peak, so as to obtain the frequency value at the resonance peak.

Images