Road roadbed disease detection method and system

Abstract

The invention discloses a road subgrade disease detection method and system, belongs to the technical field of road subgrade disease detection, realizes automatic detection and identification of road subgrade diseases by introducing a deep learning mode, and constructs a multi-scale feature extraction module and a multi-scale detection module. According to the invention, multi-scale feature recognition and detection can be carried out on the input data, so that the detection accuracy of the road subgrade diseases is further improved, the detection cost of the road subgrade diseases can be effectively saved, and the detection efficiency of the road subgrade diseases is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of road subgrade disease detection, and in particular relates to a road subgrade disease detection method and system. Background Art

[0002] At present, the detection method of road subgrade diseases is generally through manual inspection, which has certain limitations, such as low detection accuracy, slow speed and high cost, and cannot meet the needs of road maintenance and management. Summary of the invention

[0003] The present invention provides a road subgrade disease detection method and system, which are used to solve the problems of low accuracy, slow speed and high cost existing in the prior art.

[0004] In one aspect, the present invention provides a method for detecting road subgrade defects, comprising: Construct a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities; Collecting geological radar damage images corresponding to the roadbed to be detected, and preprocessing the geological radar damage images corresponding to the roadbed to be detected to obtain target geological radar damage images; The target geological radar damage image is used as input data of a road subgrade damage detection model with detection capability to obtain a road subgrade damage detection result corresponding to the road subgrade to be detected.

[0005] Furthermore, a road subgrade disease detection model is constructed, including: Construct a multi-scale feature extraction module to extract multi-scale features from the geological radar damage image input into the roadbed damage detection model through the multi-scale feature extraction module to obtain a multi-scale feature map; Constructing a multi-scale detection module to detect the multi-scale feature map extracted by the multi-scale feature extraction module through the multi-scale detection module to obtain a detection result of the region of interest; Construct a data fusion module to perform pooling and feature fusion on the detection results of the region of interest output by the multi-scale detection module to obtain fusion features; Construct a detection result output module to identify the fusion features output by the data fusion module, so as to obtain the disease monitoring results corresponding to the geological radar disease image.

[0006] Furthermore, a multi-scale feature extraction module is constructed, including: Constructing a first convolutional layer, a second convolutional layer, a first maximum pooling layer, a third convolutional layer, a fourth convolutional layer, a second maximum pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, a third maximum pooling layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a fourth maximum pooling layer, an eleventh convolutional layer, a twelfth convolutional layer, and a thirteenth convolutional layer connected in sequence; wherein the thirteenth convolutional layer outputs a first feature map, and the tenth convolutional layer outputs a second feature map; The seventh convolution layer is also connected to the seventeenth convolution layer. After the first feature map is processed by the eighteenth convolution layer and the second deconvolution layer, the feature maps output by the seventeenth convolution layer and the second deconvolution layer are added pixel by pixel through the second fusion layer to obtain a first feature addition map. After the first feature addition map is processed by the nineteenth convolution layer, a third feature map is output. The fourth convolution layer is also connected to the fourteenth convolution layer. After the second feature map is processed by the fifteenth convolution layer and the first deconvolution layer, the feature maps output by the fourteenth convolution layer and the first deconvolution layer are added pixel by pixel through the first fusion layer to obtain a second feature addition map. After the second feature addition map is processed by the sixteenth convolution layer, the fourth feature map is output; wherein the first fusion layer and the second fusion layer both adopt additive fusion functions.

[0007] Furthermore, a multi-scale detection module is constructed, including: The RPN network is used to construct a first detection network for processing the first feature map, the RPN network is used to construct a second detection network for processing the second feature map, the RPN network is used to construct a third detection network for processing the third feature map, and the RPN network is used to construct a fourth detection network for processing the fourth feature map.

[0008] Furthermore, a data fusion module is constructed, including: constructing a ROI pooling layer, a splicing layer, a first fully connected layer and a second fully connected layer that are connected in sequence; wherein the splicing layer adopts a cascade fusion function.

[0009] Furthermore, a detection result output module is constructed, including: constructing a softmax output layer for classification and a bounding box regression layer for border positioning.

[0010] Furthermore, the loss function corresponding to the roadbed disease detection model is constructed, including:

[0011]

[0012]

[0013]

[0014]

[0015] in, represents the loss function, Represents the total number of detection boxes, Indicates i The classification loss of the detection box, It indicates the probability of the existence of roadbed disease in the predicted detection box. Indicates the probability that there is a roadbed defect in the calibration detection frame, Takes 0 or 1; When it is 0, it means that there is no roadbed disease in the calibration detection frame; When the value is 1, it means that there is a roadbed defect in the calibration detection frame; represents the balance factor, represents the border loss, represents the predicted bounding box parameters, Represents the calibration frame parameters, ( ) represents the center point coordinates of the predicted bounding box, Indicates the width of the predicted border, Represents the height of the predicted border, ( ) represents the coordinates of the center point of the calibration frame, Indicates the width of the calibration border. Indicates the height of the calibration border.

[0016] Furthermore, with the goal of minimizing the loss function, the road subgrade disease detection model is trained to obtain a road subgrade disease detection model with detection capabilities, including: The maximum number of training times was set to 50,000, the basic learning rate was set to 0.006, and the Adam optimization algorithm was used to train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities. Among them, the learning rate is set to an adaptive learning rate, specifically:

[0017] in, Indicates k The learning rate during the training process, Indicates k +1 learning rate during training, represents the adjustment factor, and T represents the preset maximum number of training times.

[0018] Furthermore, after preprocessing the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained, including: after denoising and size unification processing are performed on the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained.

[0019] On the other hand, the present invention provides a road subgrade disease detection system, comprising: a model building module, an image processing module and a disease detection module; The model building module is used to build a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities; The image processing module is used to collect the geological radar damage image corresponding to the roadbed to be detected, and obtain the target geological radar damage image after preprocessing the geological radar damage image corresponding to the roadbed to be detected; The disease detection module is used to use the target geological radar disease image as input data of a road subgrade disease detection model with detection capability to obtain a road subgrade disease detection result corresponding to the road subgrade to be detected.

[0020] The present invention provides a road subgrade disease detection method and system, which realize automatic detection and identification of road subgrade diseases by introducing deep learning, and construct a multi-scale feature extraction module and a multi-scale detection module, which can perform multi-scale feature recognition and detection on input data, thereby further improving the detection accuracy and qualitative analysis of road subgrade diseases, thereby effectively saving the detection cost of road subgrade diseases and improving the detection efficiency of road subgrade diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0022] Figure 1 A flow chart of a road subgrade disease detection method provided by an embodiment of the present invention.

[0023] Figure 2 A schematic diagram of the structure of a road subgrade disease detection system provided in an embodiment of the present invention.

[0024] Figure 3 A schematic diagram of the structure of a road subgrade disease detection model provided in an embodiment of the present invention.

[0025] Figure 4 A schematic diagram of the structure of a multi-scale feature extraction module provided in an embodiment of the present invention.

[0026] Among them, 201 is a model building module, 202 is an image processing module, and 203 is a disease detection module.

[0027] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0028] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0029] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0030] like Figure 1 As shown, the present invention provides a road subgrade disease detection method, comprising: S101. Construct a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities.

[0031] Building a road subgrade disease detection model can include: (1) Construct a conventional image classification neural network; for example, construct a convolutional neural network, which can be used to classify geological radar disease images, thereby achieving the classification of a single image and realizing relatively rough roadbed disease detection.

[0032] (2) Construct a target detection neural network model with a detection box and classification probability; for example, construct a YOLOv5 neural network, which can be used to locate and identify the location of the disease in the geological radar disease image, thereby achieving more precise path disease detection.

[0033] Different types of neural network models can be selected for disease detection according to actual needs, thereby achieving roadbed disease detection with different effects.

[0034] S102, collecting geological radar damage images corresponding to the roadbed to be detected, and preprocessing the geological radar damage images corresponding to the roadbed to be detected to obtain target geological radar damage images; Collecting geological radar damage images corresponding to the roadbed to be detected can include: using an autonomous driving vehicle equipped with a geological radar to collect geological radar damage images along a preset route; after collecting the geological radar damage images, denoising and / or resizing them can be performed on them, so as to convert the geological radar damage images into images that can be recognized by the road bed damage detection model.

[0035] It is worth noting that the above-mentioned method of acquiring geological radar damage images is only an example in the embodiment of the present invention, and geological radar damage images can also be acquired by other means.

[0036] S103, using the target geological radar damage image as input data of a road subgrade damage detection model with detection capability, and obtaining a road subgrade damage detection result corresponding to the road subgrade to be detected.

[0037] After training, the road subgrade disease detection model has the ability to detect subgrade diseases, so it can directly detect the target geological radar disease images acquired subsequently, but the geological radar disease images must be resized so that they can be correctly identified.

[0038] Although existing neural networks can achieve classification, they all extract image features of a single scale and cannot effectively identify geological radar defect images, resulting in low recognition accuracy. Therefore, an embodiment of the present invention provides a road subgrade defect detection model to perform multi-scale recognition of geological radar defect images, thereby improving recognition accuracy.

[0039] like Figure 3 As shown in the figure, a road subgrade disease detection model is constructed, including: Construct a multi-scale feature extraction module to extract multi-scale features from the geological radar damage image input into the roadbed damage detection model through the multi-scale feature extraction module to obtain a multi-scale feature map; Constructing a multi-scale detection module to detect the multi-scale feature map extracted by the multi-scale feature extraction module through the multi-scale detection module to obtain a detection result of the region of interest; Construct a data fusion module to perform pooling and feature fusion on the detection results of the region of interest output by the multi-scale detection module to obtain fusion features; Construct a detection result output module to identify the fusion features output by the data fusion module, so as to obtain the disease monitoring results corresponding to the geological radar disease image.

[0040] like Figure 4As shown, a multi-scale feature extraction module is constructed, including: Constructing a first convolutional layer, a second convolutional layer, a first maximum pooling layer, a third convolutional layer, a fourth convolutional layer, a second maximum pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, a third maximum pooling layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a fourth maximum pooling layer, an eleventh convolutional layer, a twelfth convolutional layer, and a thirteenth convolutional layer connected in sequence; wherein the thirteenth convolutional layer outputs a first feature map, and the tenth convolutional layer outputs a second feature map; The seventh convolution layer is also connected to the seventeenth convolution layer. After the first feature map is processed by the eighteenth convolution layer and the second deconvolution layer, the feature maps output by the seventeenth convolution layer and the second deconvolution layer are added pixel by pixel through the second fusion layer to obtain a first feature addition map. After the first feature addition map is processed by the nineteenth convolution layer, a third feature map is output. The fourth convolution layer is also connected to the fourteenth convolution layer. After the second feature map is processed by the fifteenth convolution layer and the first deconvolution layer, the feature maps output by the fourteenth convolution layer and the first deconvolution layer are added pixel by pixel through the first fusion layer to obtain a second feature addition map. After the second feature addition map is processed by the sixteenth convolution layer, the fourth feature map is output; wherein the first fusion layer and the second fusion layer both adopt additive fusion functions.

[0041] Through the above construction process, a feature pyramid is formed to extract multi-scale features of the road subgrade, thereby making full use of features at all levels and effectively improving detection accuracy.

[0042] In the embodiment of the present invention, the working principle of the multi-scale feature extraction module is: The convolution kernel size of the first convolution layer and the second convolution layer is set to 3×3, the number of convolution kernel channels is set to 64, the step size is set to 1, and the feature map of output size is 224×224×64, where 224×224 represents the feature map size and 64 represents the number of feature map channels; The convolution kernel size of the first maximum pooling layer, the second pooling layer, the third pooling layer, and the fourth pooling layer are all set to 3×3, the step size is 1, and the padding is 1. Therefore, the image size remains unchanged after convolution, but the pooling layer will reduce the image resolution by 1 / 2.

[0043] The convolution kernel size of the third and fourth convolution layers is set to 3×3, the number of convolution kernel channels is set to 128, the step size is set to 1, and the output feature map size is 112×112×128, where 112×112 represents the feature map size and 128 represents the number of feature map channels; The convolution kernel size of the fifth, sixth, and seventh convolution layers is set to 3×3, the number of convolution kernel channels is set to 256, the step size is set to 1, and the feature map of output size is 56×56×256, where 56×56 represents the feature map size and 256 represents the number of feature map channels; The convolution kernel size of the eighth, ninth, and tenth convolutional layers is set to 3×3, the number of convolution kernel channels is set to 512, the step size is set to 1, and the feature map of output size is 28×28×512, where 28×28 represents the feature map size and 512 represents the number of feature map channels; The convolution kernel size of the eleventh, twelfth, and thirteenth convolutional layers is set to 3×3, the number of convolution kernel channels is set to 512, the step size is set to 1, and the output feature map size is 14×14×512, where 14×14 represents the feature map size and 512 represents the number of feature map channels; The first feature map changes its dimension through the eighteenth convolutional layer with a size of 1×1×256 so that the number of channels is 256, and then deconvolution is performed to make the size of the first feature map the same as the feature map size output by the seventh convolutional layer. Then, the feature map output by the seventh convolutional layer is processed by the seventeenth convolutional layer with a size of 1×1×256, and the number of feature map channels is modified to 256. Then, fusion is performed to obtain the first feature addition map. Finally, the first feature addition map is processed by the nineteenth convolutional layer with a size of 3×3×256 to remove the aliasing effect caused by upsampling.

[0044] The fusion process of the second feature map and the feature map output by the fourth convolutional layer is the same and will not be repeated here.

[0045] Through the above processing, the information of all feature layers is fully utilized, so that the RPN network can judge whether it is a foreground target more accurately, which can effectively improve the detection accuracy of small targets, thereby making the road subgrade disease detection more accurate.

[0046] In an embodiment of the present invention, a multi-scale detection module is constructed, including: The RPN (Region Proposal Network) network is used to construct a first detection network for processing the first feature map, the RPN network is used to construct a second detection network for processing the second feature map, the RPN network is used to construct a third detection network for processing the third feature map, and the RPN network is used to construct a fourth detection network for processing the fourth feature map.

[0047] In an embodiment of the present invention, constructing a data fusion module includes: constructing a ROI pooling layer, a splicing layer, a first fully connected layer and a second fully connected layer connected in sequence; wherein the splicing layer adopts a cascade fusion function.

[0048] In an embodiment of the present invention, constructing a detection result output module includes: constructing a softmax output layer for classification and a bounding box regression layer (bounding box regression layer) for border positioning.

[0049] The region of interest is input into the ROI pooling layer and mapped back to the corresponding pyramid feature layer to obtain a series of feature blocks of the region of interest with a uniform size of 7×7. Finally, the cascade fusion function is used to vertically splice the feature maps of multiple levels together to share the classification layer and regression layer.

[0050] In an embodiment of the present invention, the loss function corresponding to the roadbed disease detection model is constructed, including:

[0051]

[0052]

[0053]

[0054]

[0055] in, represents the loss function, Represents the total number of detection boxes, Indicates i The classification loss of the detection box, It indicates the probability of the existence of roadbed disease in the predicted detection box. Indicates the probability that there is a roadbed defect in the calibration detection frame, Takes 0 or 1; When it is 0, it means that there is no roadbed disease in the calibration detection frame; When the value is 1, it means that there is a roadbed defect in the calibration detection frame; represents the balance factor, represents the border loss, represents the predicted bounding box parameters, Represents the calibration frame parameters, ( ) represents the center point coordinates of the predicted bounding box, Indicates the width of the predicted border, Represents the height of the predicted border, ( ) represents the coordinates of the center point of the calibration frame, Indicates the width of the calibration border. Indicates the height of the calibration border.

[0056] It is worth noting that the above loss function is only a preferred implementation method provided by the embodiment of the present invention, and other loss functions can also be used to train the road subgrade disease detection model.

[0057] In the embodiment of the present invention, the road subgrade disease detection model is trained with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capability, including: The maximum number of training times was set to 50,000, the basic learning rate was set to 0.006, and the Adam optimization algorithm was used to train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities. Among them, the learning rate is set to an adaptive learning rate, specifically:

[0058] in, Indicates k The learning rate during the training process, Indicates k +1 learning rate during training, represents the adjustment factor, and T represents the preset maximum number of training times.

[0059] The present invention provides an adaptive learning rate, which can realize variable-speed learning, with a relatively gentle early stage and a large drop in the later stage, thereby improving the refinement of training.

[0060] In an embodiment of the present invention, after preprocessing the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained, including: after denoising and size unification processing are performed on the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained.

[0061] Alternatively, the geological radar damage images may simply be resized so that the constructed roadbed damage detection model can identify and process them.

[0062] like Figure 2 As shown, the present invention provides a road subgrade disease detection system, comprising: a model building module 201, an image processing module 202 and a disease detection module 203; The model building module 201 is used to build a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities; The image processing module 202 is used to collect the geological radar damage image corresponding to the roadbed to be detected, and obtain the target geological radar damage image after preprocessing the geological radar damage image corresponding to the roadbed to be detected; The disease detection module 203 is used to use the target geological radar disease image as input data of a road subgrade disease detection model with detection capability to obtain a road subgrade disease detection result corresponding to the road subgrade to be detected.

[0063] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The solutions in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and interpreted scripting language JavaScript, etc.

[0064] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0065] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0066] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0067] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0068] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A road subgrade disease detection method, characterized in that: include: Construct a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities; Collecting geological radar damage images corresponding to the roadbed to be detected, and preprocessing the geological radar damage images corresponding to the roadbed to be detected to obtain target geological radar damage images; The target geological radar damage image is used as input data of a road subgrade damage detection model with detection capability to obtain a road subgrade damage detection result corresponding to the road subgrade to be detected.

2. The road subgrade disease detection method according to claim 1, characterized in that: Construct a road subgrade disease detection model, including: Construct a multi-scale feature extraction module to extract multi-scale features from the geological radar damage image input into the road subgrade damage detection model through the multi-scale feature extraction module to obtain a multi-scale feature map; Constructing a multi-scale detection module to detect the multi-scale feature map extracted by the multi-scale feature extraction module through the multi-scale detection module to obtain a detection result of the region of interest; Construct a data fusion module to perform pooling and feature fusion on the detection results of the region of interest output by the multi-scale detection module to obtain fusion features; Construct a detection result output module to identify the fusion features output by the data fusion module, so as to obtain the disease monitoring results corresponding to the geological radar disease image.

3. The road subgrade disease detection method according to claim 2, characterized in that: Construct a multi-scale feature extraction module, including: Constructing a first convolutional layer, a second convolutional layer, a first maximum pooling layer, a third convolutional layer, a fourth convolutional layer, a second maximum pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, a third maximum pooling layer, an eighth convolutional layer, a ninth convolutional layer, a tenth convolutional layer, a fourth maximum pooling layer, an eleventh convolutional layer, a twelfth convolutional layer, and a thirteenth convolutional layer connected in sequence; wherein the thirteenth convolutional layer outputs a first feature map, and the tenth convolutional layer outputs a second feature map; The seventh convolution layer is also connected to the seventeenth convolution layer. After the first feature map is processed by the eighteenth convolution layer and the second deconvolution layer, the feature maps output by the seventeenth convolution layer and the second deconvolution layer are added pixel by pixel through the second fusion layer to obtain a first feature addition map. After the first feature addition map is processed by the nineteenth convolution layer, a third feature map is output. The fourth convolution layer is also connected to the fourteenth convolution layer. After the second feature map is processed by the fifteenth convolution layer and the first deconvolution layer, the feature maps output by the fourteenth convolution layer and the first deconvolution layer are added pixel by pixel through the first fusion layer to obtain a second feature addition map. After the second feature addition map is processed by the sixteenth convolution layer, the fourth feature map is output; wherein the first fusion layer and the second fusion layer both adopt additive fusion functions.

4. The road subgrade disease detection method according to claim 3, characterized in that: Construct a multi-scale detection module, including: The RPN network is used to construct a first detection network for processing the first feature map, the RPN network is used to construct a second detection network for processing the second feature map, the RPN network is used to construct a third detection network for processing the third feature map, and the RPN network is used to construct a fourth detection network for processing the fourth feature map.

5. The road subgrade disease detection method according to claim 4, characterized in that: Constructing a data fusion module includes: constructing a ROI pooling layer, a splicing layer, a first fully connected layer, and a second fully connected layer that are connected in sequence; wherein the splicing layer adopts a cascade fusion function.

6. The road subgrade disease detection method according to claim 5, characterized in that: Construct the detection result output module, including: constructing a softmax output layer for classification and a bounding box regression layer for border positioning.

7. The road subgrade disease detection method according to claim 6, characterized in that: The loss function corresponding to the road subgrade disease detection model is constructed, including: in, represents the loss function, Represents the total number of detection boxes, Indicates i The classification loss of the detection box, It indicates the probability of the existence of roadbed disease in the predicted detection box. Indicates the probability that there is a roadbed defect in the calibration detection frame, Takes 0 or 1; When it is 0, it means that there is no roadbed disease in the calibration detection frame; When the value is 1, it means that there is a roadbed defect in the calibration detection frame; represents the balance factor, represents the border loss, represents the predicted bounding box parameters, Represents the calibration frame parameters, ( ) represents the center point coordinates of the predicted bounding box, Indicates the width of the predicted border, Represents the height of the predicted border, ( ) represents the coordinates of the center point of the calibration frame, Indicates the width of the calibration border. Indicates the height of the calibration border.

8. The road subgrade disease detection method according to claim 7, characterized in that: With the goal of minimizing the loss function, the road subgrade disease detection model is trained to obtain a road subgrade disease detection model with detection capabilities, including: The maximum number of training times was set to 50,000, the basic learning rate was set to 0.006, and the Adam optimization algorithm was used to train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities. Among them, the learning rate is set to an adaptive learning rate, specifically: in, Indicates k The learning rate during the training process, Indicates k +1 learning rate during training, represents the adjustment factor, and T represents the preset maximum number of training times.

9. The road subgrade disease detection method according to claim 8, characterized in that: After preprocessing the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained, including: after denoising and size unification processing are performed on the geological radar damage image corresponding to the roadbed to be detected, a target geological radar damage image is obtained.

10. A road subgrade disease detection system, characterized in that: include: Model building module, image processing module and disease detection module; The model building module is used to build a road subgrade disease detection model and a loss function corresponding to the road subgrade disease detection model, and train the road subgrade disease detection model with the goal of minimizing the loss function to obtain a road subgrade disease detection model with detection capabilities; The image processing module is used to collect the geological radar damage image corresponding to the roadbed to be detected, and obtain the target geological radar damage image after preprocessing the geological radar damage image corresponding to the roadbed to be detected; The disease detection module is used to use the target geological radar disease image as input data of a road subgrade disease detection model with detection capability to obtain a road subgrade disease detection result corresponding to the road subgrade to be detected.

Images