Intelligent termite detection system based on MIMO radar technology and deep learning

The intelligent termite detection system uses MIMO radar and deep learning to enhance termite activity monitoring accuracy and real-time performance, addressing the limitations of existing systems by integrating a box device with a Raspberry Pi unit and MobileNetV2 model for precise and timely detection on mobile devices.

CN119087430AActive Publication Date: 2024-12-06UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411375458.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-30
Publication Date
2024-12-06
Estimated Expiration
2044-09-30

AI Technical Summary

Technical Problem

Current termite detection systems lack precise positioning and real-time monitoring capabilities, especially for termite activities hidden in walls, and have limitations in data processing complexity and adaptability to mobile platforms.

Method used

An intelligent termite detection system combining MIMO radar technology with deep learning, utilizing a box device and mobile terminal, where the box device includes a microwave radar and Raspberry Pi data processing unit to collect and preprocess radar echo signals, and a MobileNetV2 model on the mobile terminal for real-time termite activity monitoring and visualization.

Benefits of technology

The system achieves high accuracy and real-time detection with a delay of 0.2 seconds, improving detection accuracy from 87.6% to 95.0% and ensuring reliable, rapid response to termite activities with user-friendly mobile implementation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119087430A_ABST
    Figure CN119087430A_ABST
Patent Text Reader

Abstract

The invention discloses an intelligent termite detection system based on an MIMO radar technology and deep learning, and relates to the technical field of target identification. The system comprises box equipment and a mobile terminal, the box equipment is used for being deployed near a detection target, radar detection is carried out on the detection target through a microwave radar, a radar echo signal in the detection target is obtained, and corresponding radar imaging and radar energy are wirelessly transmitted to the mobile terminal; and the mobile terminal carries out termite identification based on the deployed detection model, and determines the activity area of the termites based on the radar energy change. By combining the MIMO radar technology, the deep learning algorithm and radar energy change detection, the termite activity detection accuracy is remarkably improved, the real-time performance is high, the detection program of the mobile terminal is high in adaptability and user friendliness, implementation of termite detection is facilitated, convenience is provided for follow-up processing, and the termite activity detection method is suitable for popularization and application. A user can check a detection result on the mobile terminal and take corresponding prevention and control measures according to the result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of target recognition technology, and in particular to an intelligent termite detection system based on MIMO radar technology and deep learning. Background Art

[0002] Termites are social insects that usually live underground or inside buildings in highly organized colonies. Termites feed on wood and cellulose and can damage building structures, wooden furniture and other materials containing cellulose. Due to their concealment and efficient reproduction ability, termite activities are often difficult to detect early, resulting in serious economic losses and safety hazards.

[0003] The damage caused by termites mainly manifests itself in the following aspects:

[0004] Building damage: Termites eat away at wooden structures, weakening the stability of buildings, and in severe cases may cause them to collapse.

[0005] Economic losses: Termites are highly destructive and repairing eroded buildings and furniture requires a large amount of capital investment.

[0006] Safety hazards: Buildings and furniture infested by termites pose safety hazards and may pose a threat to the personal safety of occupants.

[0007] Environmental impact: Termite activities change soil structure, affect plant growth, and thus have an impact on the ecological environment.

[0008] Current termite detection methods have significant deficiencies in positioning accuracy and operational efficiency, especially for termite activities hidden in walls. Traditional detection methods rely on manual experience and are difficult to accurately detect hidden areas. In recent years, MIMO radar technology has provided a new technical approach for termite detection with its excellent penetration ability and high resolution. Combined with deep learning algorithms, the accuracy and efficiency of detection can be further improved.

[0009] Existing termite detection systems mostly use manual or semi-automatic methods, which have limitations in precise positioning and real-time monitoring and cannot meet actual needs. Since radar signals have penetrating power and can detect objects hidden inside walls, combined with deep learning technology, accurate identification and monitoring of termite activities can be achieved. However, existing solutions still have many technical defects in practical applications, such as complex data processing, poor real-time performance, and insufficient adaptability to mobile platforms. . Summary of the invention

[0010] The purpose of the present invention is to provide an intelligent termite detection system based on MIMO radar technology and deep learning, which realizes real-time monitoring and accurate identification of termite activities by combining microwave radar and deep learning model.

[0011] The technical solution adopted by the present invention is:

[0012] An intelligent termite detection system based on MIMO radar technology and deep learning, which includes a box device and a mobile terminal;

[0013] The box device is used to be deployed near the detection target, and detects the detection target through microwave radar to obtain the radar echo signal inside the detection target;

[0014] The box device includes an outer box, an inner box, a microwave radar device, a Raspberry Pi data processing unit, a power module and an external interface; wherein the microwave radar device is fixed at the bottom of the outer box, and its radar scanning direction is downward; the Raspberry Pi data processing unit connected to the microwave radar device is placed on the top of the outer box, the power module is used to supply power to the microwave radar device and the Raspberry Pi data processing unit, and the outer box of the microwave radar device is also configured with an external interface, which is used as a power interface and a data transmission interface;

[0015] The Raspberry Pi data processing unit is used to set the working parameters of the microwave radar device. The microwave radar device is used to collect the radar echo signal inside the detection target and send it to the Raspberry Pi data processing unit; the Raspberry Pi data processing unit is used to filter the radar echo signal for noise, and then obtain the corresponding two-dimensional image data through imaging processing and perform image enhancement processing on it to obtain the data to be detected; the data to be detected is transmitted to the mobile terminal via the wireless network; the Raspberry Pi data processing unit counts the radar energy of each area based on the received echo signal according to the division form of the image area generated by the preset echo signal, and transmits it to the mobile terminal via the wireless network;

[0016] The mobile terminal is equipped with a termite detection application based on the MobileNetV2 model, which is used to perform target detection on the received data to be detected and obtain preliminary detection results; and based on the monitoring of the changes in radar echo energy, the activity area of ​​​​the termites in the preliminary detection results is confirmed, and the final termite detection results are generated and visualized on the mobile terminal.

[0017] That is, the present invention is based on a box device deployed near the detection target (such as a cement wall, a wooden wall, or other wall targets), a remote mobile terminal, and a preset user interaction program and a MobileNetV2 deep learning model to achieve real-time monitoring of the detection target and visualization of termite activities.

[0018] Furthermore, the user interface of the termite detection application on the mobile terminal includes:

[0019] Start / Stop function: used to start and stop data saving. A "Start / Stop" operation item can be set at the top of the user interface of the mobile terminal so that users can start or stop the visualization process at any time to avoid the generation of redundant data.

[0020] Real-time monitoring display area: used to display real-time video or images transmitted by microwave radar equipment, and to mark the confidence level of the corresponding area when termite activity is detected;

[0021] Data saving function: A “Save” operation item can be set at the top of the user interface, and users can save relevant data after the test is completed for subsequent analysis.

[0022] Furthermore, the user interface of the termite detection application in the mobile terminal also includes a historical detection record query function for users to view and analyze past detection data, which supports filtering and reviewing by time conditions.

[0023] Furthermore, the noise filtering uses a high-pass filter to remove background noise signals that do not exceed a specified frequency.

[0024] Furthermore, the Raspberry Pi data processing unit performs image enhancement processing using a histogram equalization method.

[0025] Furthermore, the MobileNetV2 model sequentially includes an input layer, a starting convolutional layer, an inverted residual block sequence, a convolutional layer before the end, a local average pooling layer, and a fully connected layer based on the Softmax activation function, which is used to output the termite detection confidence of each image area of ​​the data to be detected to obtain preliminary detection results.

[0026] Furthermore, the MobileNetV2 model is a model based on the channel attention mechanism.

[0027] Furthermore, the termite dataset required for training the MobileNetV2 model is constructed as follows:

[0028] Obtain a variety of termite-eaten wood samples;

[0029] Place a wall material with uniform thickness in the constructed experimental environment, and set up two areas behind the wall material, one area is for placing normal wood block materials, and the other area is for placing the obtained wood samples that have been eaten (hereinafter referred to as eaten samples);

[0030] The two areas are separated by a partition, and a certain number of termites, such as white-green termites, are added to the side where the infested wood sample is placed; the area division and marking are performed on the wall material with uniform thickness to correspond to the image division of the two-dimensional image obtained by imaging the radar echo signal;

[0031] The box device of the present invention is deployed in front of a wall material with uniform thickness to collect the termite data set required for MobileNetV2 model training;

[0032] Among them, the wall material with uniform thickness can be a cement wall segment or a wooden wall to simulate different detection scenarios. The model trained with the termite data set corresponding to different wall materials is used to monitor the detection targets of different wall materials. The trained MobileNetV2 models corresponding to different wall materials can be deployed on the mobile terminal, and optional detection operations for the target material can be provided to the user.

[0033] Furthermore, the mobile terminal confirms that the activity area of ​​termites in the preliminary detection result specifically includes:

[0034] Count the mean energy value of each image region within a period of time and calculate its variance as the evaluation value of each region;

[0035] Based on the set energy threshold, the confidence value of the area below the energy threshold is directly set to 0;

[0036] The area with non-zero confidence value is the activity area of ​​termites.

[0037] Furthermore, the energy threshold is set using an adaptive adjustment strategy:

[0038] Initialize the energy threshold, and based on the initial value, process whether the confidence value of each area is set to 0;

[0039] From the area where the confidence value is 0, the mean value μ of the radar echo energy is calculated b and standard deviation σ b ;

[0040] Based on the formula T adaptive =μ b +k·σ b Get the adjusted energy threshold T adaptive , where k is an adjustment coefficient used to control the strictness of the threshold, a preset value;

[0041] Furthermore, during operation, the system continuously monitors the accuracy of detecting termite activity using the adjusted energy threshold to evaluate the detection effect of the adjustment coefficient k with different values, and selects the adjustment coefficient k for the next time period based on the detection effect.

[0042] The technical solution provided by the present invention brings at least the following beneficial effects:

[0043] Improved detection accuracy: By combining MIMO radar technology, deep learning algorithms, and radar energy change detection, the accuracy of termite activity detection has been significantly improved. The initial accuracy of this system using the MobileNet model was 87.6%, and after combining radar energy change detection, the accuracy increased to 95.0%. This significant improvement in the technical solution has greatly enhanced the reliability and practicality of the system in practical applications.

[0044] Strong real-time performance: Raspberry Pi and mobile terminals are used for data processing and model reasoning to achieve real-time detection and feedback. The entire process from scanning data to obtaining termite detection results has a delay of only 0.2 seconds. This high real-time performance ensures that the system can provide accurate detection results in a short time, which is of great significance for rapid response and taking corresponding measures.

[0045] High adaptability and user-friendliness: The system design can be implemented based on the Android platform and has good adaptability to mobile devices. Users can easily use the system through a well-designed interactive interface and record the detection location. This not only facilitates the implementation of termite detection, but also provides convenience for subsequent treatment (such as termite killing, etc.). Users can view the detection results on their mobile phones and take corresponding prevention and control measures based on the results. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0047] Figure 1 The structure of the intelligent termite detection system based on MIMO radar technology and deep learning is a meaningful diagram.

[0048] Figure 2 Schematic diagram of the main structure of the intelligent termite detection system based on MIMO radar technology and deep learning.

[0049] Figure 3 Workflow diagram for termite detection.

[0050] Figure 4 Flowchart for data processing and model training.

[0051] Figure 5 This is a schematic diagram of the mobile terminal application interface.

[0052] Among them, 1 is the inner box, 2 is the microwave radar equipment, 3 is the Raspberry Pi data processing unit, 4 is the outer box, 5 is the power module, and 6 is the external interface. DETAILED DESCRIPTION

[0053] To make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in detail and completely in conjunction with the drawings in the implementation of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings can be arranged and designed using different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the present application for protection, but merely represents selected embodiments of the present invention.

[0054] The intelligent termite detection system based on MIMO radar technology and deep learning provided by the embodiment of the present invention includes a box device and a mobile terminal, wherein the box device is used to be deployed near the detection target, such as Figure 1 As shown, it includes an outer box 4, an inner box 1, a microwave radar device 2, a Raspberry Pi data processing unit 3, a power module 5 and an external interface 6 (such as a charging interface, a USB interface, etc.). The external interface is located on one side of the outer box 4 to facilitate the connection and charging of the system; the microwave radar device 2 is fixed to the bottom of the outer box 1, and the scanning direction of the radar is downward; the Raspberry Pi data processing unit 3 and the power module 5 are respectively located in the middle and one side of the inner box to provide data processing and power support for the stable operation of the system. To express the structure clearly, Figure 1 The outer box 4 is made transparent, that is, the box 4 only shows its outline without hindering the display of the internal structure of the box, so as to display the microwave radar device 2, Raspberry Pi data processing unit 3, and power module 5 hidden in the box. Figure 2 It is a schematic diagram of the main structure of the box device, which shows the layout relationship of the internal components, specifically: the outer box 4 contains the internal microwave radar device 2, the Raspberry Pi data processing unit 3 and the power module 5. In this embodiment, the microwave radar device 2 is fixed to the bottom of the inner box 1, and relative to the position of the outer box 4, the microwave radar device 2 is connected to the Raspberry Pi data processing unit 3 through a USB connection, and the power module 5 provides stable power for both.

[0055] In an embodiment of the present invention, the microwave radar device 2 is used to perform radar scanning on the detection target. It adopts a MIMO antenna array structure, can obtain radar echo data of the scanned area and obtain 2D image data through an imaging algorithm, and transmits it to the Raspberry Pi data processing unit 3 via a USB connection for data preprocessing (noise filtering and image enhancement), obtains the data to be detected, and transmits it to the mobile terminal via a wireless network.

[0056] The mobile terminal is equipped with a termite detection application based on the MobileNetV2 model, which is used to perform target detection on the received data to be detected and obtain preliminary detection results; and based on the monitoring of the radar echo energy changes, the activity area of ​​the termites in the preliminary detection results is confirmed, and the final termite detection results are generated and visualized on the mobile terminal. Furthermore, the final termite detection results can be pushed to the designated user for monitoring and alarm. The embodiment of the present invention can more accurately confirm the existence of termite activities by monitoring the radar energy changes, because the tiny movement of termites will cause energy fluctuations, and the accuracy of termite detection can be effectively improved by combining the energy criterion.

[0057] Typically, the MobileNetV2 model specifically includes: an input layer for receiving input images, i.e., preprocessed 2D image data; a starting convolution layer, a standard 3×3 convolution layer, for preliminary feature extraction; a sequence of inverted residual blocks, which consists of multiple inverted residual blocks, and a convolution layer before the end: a 1x1 point convolution for generating the final feature map; a global average pooling layer: compressing the feature map into a single vector; a fully connected layer: for the output of the classification task. Among them, each inverted residual block includes: an expansion 1x1 convolution (expand convolution) for increasing the number of channels; a depth convolution 3x3 or 5x5, whose step size can be set to 1 or 2, for spatial feature extraction; a linear bottleneck 1x1 convolution (squeeze and excite), i.e., a depth-separable convolution, for dimensionality reduction; a jump connection (only when the input and output dimensions are the same); and a fully connected layer (also called an output layer) set as a fully connected layer with a Softmax activation function, which is used to generate a probability distribution of the binary category (whether termites exist) for each termite detection, thereby outputting a confidence value. These confidence values ​​can be used to evaluate the model's confidence level for each prediction result, allowing users to set judgment thresholds based on specific needs.

[0058] Figure 3The complete workflow of the detection system provided by the embodiment of the present invention for detecting termites is described: after the system is started, the microwave radar device 2 is initialized by connecting to the Raspberry Pi data processing unit 3, setting radar parameters and establishing a data transmission link. After the initialization is completed, the microwave radar device 2 starts to scan the target area, collects radar signals and 2D image data in the wall, and transmits the collected data to the Raspberry Pi in the form of a real-time stream for preprocessing. In this process, the system further monitors the radar energy changes. The tiny movement of termites will cause significant changes in the energy value received by the radar. By analyzing these energy changes, the system can more accurately confirm termite activity. The data after energy change detection is input into the MobileNetV2 model for inference. The model classifies the data, determines whether there is termite activity, and outputs the detection results. The detection results are transmitted to the user's mobile phone via a wireless network. The user can view the detection results through the mobile phone application and obtain detailed visualization information of the location and degree of termite activity. The user can also take further termite prevention and control measures based on the detection report provided by the application, such as notifying a professional team to handle it. In addition, the application supports recording and storing detection history for user query and review.

[0059] In the embodiment of the present invention, see Figure 4 The specific implementation process of the MobileNetV2 model used in model training includes:

[0060] First, the target area is scanned using radar equipment. The collected radar signals and 2D image data will go through a series of preprocessing steps. To ensure data quality, a Butterworth high-pass filter is used to remove low-frequency noise (such as baseline drift below 0.1Hz), and histogram equalization is used to enhance the contrast of the image, making the subtle changes in termite activity more clearly visible. In the preprocessing stage, data enhancement techniques such as contrast enhancement, brightness enhancement, image flipping and rotation are also applied according to the living environment conditions of termites to improve the model's adaptability to diverse data.

[0061] During the data annotation stage, the open source tool LabelImg is used to manually annotate the collected image data to ensure accurate identification of termite activity areas. The radar imaging area is divided into multiple area blocks (for example, 18 area blocks) for annotation to more accurately locate the position of termites. If there is enough annotated data, a pre-trained model can be used for automatic annotation, and manual inspection and adjustment can be performed to ensure the accuracy of the annotation. To prevent the occurrence of data imbalance problems, this embodiment divides the data set into a training set, a validation set, and a test set in a ratio of 7:1:2 to ensure that the samples of each subset are evenly distributed. The data set is divided into two types, collected on cement walls and wooden walls, which are used to train two models for users to choose according to actual conditions.

[0062] In terms of the choice of model architecture, MobileNetV2 is used as the basic model. Its deep separable convolution significantly reduces the amount of computation and is very suitable for running on resource-constrained devices. In order to further improve the feature extraction capability of the model, the Coordinate Attention (CA) mechanism is introduced, which can effectively capture the information of different spatial positions in the feature map, thereby improving the accuracy of termite activity detection. You can choose to add a channel attention mechanism module in the middle of each inverted residual block of the MobileNetV2 model, such as adding an SE module, which includes global average pooling and two fully connected layers. The global average pooling is used to extract the importance of each channel in the feature map, and then the weights between channels are learned through the fully connected layer. Finally, the feature maps of different channels are weighted fused to achieve the regulation of channel attention.

[0063] In terms of hyperparameter settings, the initial learning rate is set to 0.001, and a learning rate scheduler such as ReduceLROnPlateau is used to automatically adjust the learning rate according to the validation set loss to prevent overfitting. The batch size (BatchSize) is set to 32 to ensure that the gradient estimation is more stable. In addition, during the model training process, the Adam optimizer is used to accelerate the convergence of the model, and data enhancement techniques such as rotation, translation, and scaling are applied to further improve the generalization ability of the model.

[0064] In terms of model evaluation, we use indicators such as accuracy, precision, recall, and F1 score for comprehensive evaluation, paying special attention to the number of false positive and false negative samples in the confusion matrix to ensure that the model's classification ability for borderline samples is optimized. We introduce k-fold cross validation to verify the stability and generalization ability of the model. In the later stages of training, we further optimize the model weights by gradually reducing the learning rate, and combine the early stopping mechanism to prevent overfitting.

[0065] In the embodiment of the present invention, the termite dataset is constructed as follows:

[0066] By establishing contact with termite control agencies, we can obtain samples of wood eaten by termites in various forms. We use a transparent acrylic box of appropriate size as the basis for the construction of the experimental environment, and stick a wall material of uniform thickness on the front of the box. Then we divide it into two areas, one for normal wood blocks and the other for wood samples eaten by termites, separated by a partition in the middle. We add an appropriate amount of white termites to one side of the eaten sample. At the same time, we replace the wall material with cement wall sections or wooden walls during the experiment to simulate different detection scenarios. In addition to collecting data in the constructed environment, we can also collect actual data with the help of houses infested by termites. And we can divide and mark the areas on the front wall so that we can mark the model training data set later.

[0067] After the MobileNetV2 model for termite detection is trained, it is deployed in a mobile terminal to realize termite detection on the wall to be detected.

[0068] After the model detection is completed, the system further uses the energy threshold algorithm to assist in judging the detection results. Specifically, the radar imaging results are divided into 18 area blocks, and the system returns the energy value of each area through the relevant functions of the radar board. Subsequently, the system calculates the average value of the energy value of each area over a period of time and finds its variance as the evaluation value of the area. Since experiments have shown that the change in energy value can reflect whether there is biological activity in the area, the system sets an initial threshold value T0, which can be set based on the actual application scenario. In this embodiment, it is set to 10. For areas below this threshold, the confidence value is directly set to 0.

[0069] In addition, in order to eliminate the interference of environmental factors on judgment, the system will dynamically adjust the energy threshold. The specific steps include:

[0070] (1) Termite-free area assessment: The system selects those areas judged as “no termite activity” from the model detection results. These areas are considered to have no biological activity and their energy values ​​are mainly affected by environmental factors. Then, the system collects statistics on the energy assessment values ​​of these termite-free areas and calculates their mean μ b and standard deviation σ b , to reflect the normal energy fluctuation range under the environmental background.

[0071] (2) Adaptive threshold adjustment: The system dynamically adjusts the initial energy threshold T0 according to the background energy statistics. The adjusted threshold T adaptive Calculated by the following formula:

[0072] T adaptive =μ b +k·σ b

[0073] Where k is an adjustment factor used to control the strictness of the threshold. The system selects an appropriate k value based on actual application requirements. For example, k can be set to 1 or 2. Then, the system uses the adjusted threshold T adaptive All region blocks are re-evaluated. If the energy evaluation value of a region is lower than the new threshold, the confidence of the region is set to 0.

[0074] (3) Verification and feedback: During operation, the system continuously monitors the impact of the adjusted threshold on the accuracy of termite activity detection, and verifies the effects of different k values ​​based on experiments, and selects the optimal threshold adjustment strategy to balance the false detection rate and missed detection rate. The system will also iteratively optimize the threshold adjustment strategy based on the actual detection results to ensure high accuracy and low false alarm rate under various environmental conditions.

[0075] The above processing process analyzes the energy value of the termite-free area and adjusts the detection threshold in real time, so that the system can dynamically adapt to different environmental backgrounds, thereby achieving more accurate and reliable biological activity detection.

[0076] During the model deployment phase, the model quantization processing (using TensorFlow Lite) corresponding to the two materials was performed to reduce the size of the model and speed up the inference speed, ensuring real-time and high efficiency on mobile devices. Ultimately, after these fine-tuning and optimizations, the termite detection system not only improved the accuracy of detection, but also output the confidence level of each detection result, making the system more reliable and interpretable under different environmental conditions.

[0077] Figure 5 This is a schematic diagram of the termite detection application interface on the mobile terminal of an embodiment of the present invention, which shows the main operation interface of the user when using the intelligent termite detection system. In this interface, a title bar is displayed at the top, with a "Start / Stop" control button on the left and a "Save" button on the right, which is convenient for the user to save the current detection frame during the detection operation. In the middle of the interface is a large display area for real-time display of detection results. The system uses radar image results to render the current area in real time to obtain the detection results of the model in 18 area blocks, and outputs the confidence of each area. The information bar below displays the model's inference time and the wall material tab, and the "History" button allows the user to view past saved data. The design of the entire interface is concise and clear, with user experience as the center, ensuring that users can easily find the required functions during operation. Through Figure 5 , users can clearly understand the usage and operation process of the intelligent termite detection system, so as to carry out termite detection work more effectively.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0079] The above are only some embodiments of the present invention. For those skilled in the art, several modifications and improvements can be made without departing from the creative concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. Intelligent termite detection system based on MIMO radar technology and deep learning, characterized by: Including box equipment and mobile terminals; The box device is used to be deployed near the detection target, and detects the detection target through microwave radar to obtain the radar echo signal inside the detection target; The box device includes an outer box, an inner box, a microwave radar device, a Raspberry Pi data processing unit, a power module and an external interface; wherein the microwave radar device is fixed at the bottom of the outer box, and its radar scanning direction is downward; the Raspberry Pi data processing unit connected to the microwave radar device is placed on the top of the outer box, the power module is used to supply power to the microwave radar device and the Raspberry Pi data processing unit, and the outer box of the microwave radar device is also configured with an external interface, which is used as a power interface and a data transmission interface; The Raspberry Pi data processing unit is used to set the working parameters of the microwave radar device. The microwave radar device is used to collect the radar echo signal inside the detection target and send it to the Raspberry Pi data processing unit; the Raspberry Pi data processing unit is used to filter the radar echo signal for noise, and then obtain the corresponding two-dimensional image data through imaging processing and perform image enhancement processing on it to obtain the data to be detected; the data to be detected is transmitted to the mobile terminal via the wireless network; the Raspberry Pi data processing unit counts the radar energy of each area based on the received echo signal according to the division form of the image area generated by the preset echo signal, and transmits it to the mobile terminal via the wireless network; The mobile terminal is equipped with a termite detection application based on the MobileNetV2 model, which is used to perform target detection on the received data to be detected and obtain preliminary detection results; and based on the monitoring of the changes in radar echo energy, the activity area of ​​​​the termites in the preliminary detection results is confirmed, and the final termite detection results are generated and visualized on the mobile terminal.

2. The system according to claim 1, characterized in that The user interface of the termite detection application on the mobile terminal includes: Start and stop function: used to start and stop data saving; Real-time monitoring display area: used to display real-time video or images transmitted by microwave radar equipment, and to mark the confidence level of the corresponding area when termite activity is detected; Data saving function: used to save the test data and test results after the test is completed.

3. The system according to claim 2, characterized in that The user interface of the termite detection application on the mobile terminal also includes a historical detection record query function for users to view and analyze past detection data, which supports filtering and reviewing by time conditions.

4. The system according to claim 1, characterized in that Noise filtering uses a high-pass filter to remove background noise signals that do not exceed a specified frequency.

5. The system according to claim 1, wherein: The Raspberry Pi data processing unit uses histogram equalization to perform image enhancement processing.

6. The system according to claim 1, wherein: The MobileNetV2 model includes an input layer, a starting convolutional layer, an inverted residual block sequence, a convolutional layer before the end, a local average pooling layer, and a fully connected layer based on the Softmax activation function, which is used to output the termite detection confidence of each image area of ​​the data to be detected and obtain preliminary detection results.

7. The system according to claim 1, characterized in that The termite dataset required for training the MobileNetV2 model is constructed as follows: Obtain a variety of termite-eaten wood samples; Place a wall material with uniform thickness in the constructed experimental environment, and set up two areas behind the wall material, one area is for placing normal wood block materials, and the other area is for placing the obtained wood samples that have been eaten; The two areas are separated by a partition, and a certain number of termites, such as white-green termites, are added to the side where the infested wood sample is placed; the area division and marking are performed on the wall material with uniform thickness to correspond to the image division of the two-dimensional image obtained by imaging the radar echo signal; Deploy the box device in front of a wall material with uniform thickness to collect the termite dataset required for MobileNetV2 model training; Among them, the MobileNetV2 model trained with the termite data set corresponding to different wall materials is used to monitor the detection targets of different wall materials. The trained MobileNetV2 models corresponding to different wall materials are deployed on the mobile terminal, and optional detection operations for the target material are provided to the user.

8. The system of claim 1, wherein: The mobile terminal monitors the change in radar echo energy and confirms that the activity areas of termites in the preliminary detection results specifically include: Count the mean energy value of each image region within a period of time and calculate its variance as the evaluation value of each region; Based on the set energy threshold, the confidence value of the area below the energy threshold is directly set to 0; The area with non-zero confidence value is the activity area of ​​termites.

9. The system according to claim 8, characterized in that The energy threshold is set using an adaptive adjustment strategy: Initialize the energy threshold, and based on the initial value, process whether the confidence value of each area is set to 0; From the area where the confidence value is 0, the mean value μ of the radar echo energy is calculated b and standard deviation σ b ; Based on the formula T adaptive =μ b +k·σ b Get the adjusted energy threshold T adaptive , where k is an adjustment factor used to control the strictness of the threshold.

10. The system according to claim 9, characterized in that During operation, the system continuously monitors the accuracy of termite activity detection using the adjusted energy threshold to evaluate the detection effect of different values ​​of the adjustment coefficient k, and selects the adjustment coefficient k for the next time period based on the detection effect.

Citation Information

Patent Citations

  • Radar RD image target detection method under low signal-to-noise ratio based on deep learning

    CN113887583A

  • Unmanned aerial vehicle MIMO radar clutter suppression method

    CN115685094A

  • Group target detection method based on visible light remote sensing image

    CN117935075A

  • A milli meter (MM) wave imaging system for non-destructive testing and deploying methods thereof

    WO2023211360A1