Vehicle body anomaly detection method based on deep learning sample detection algorithm

Abstract

The invention belongs to the technical field of vehicle body detection. The invention provides a vehicle body anomaly detection method based on a deep learning sample detection algorithm. The method comprises the following steps: acquiring first appearance image data of a train body, and preprocessing the first appearance image data; calling a corresponding train body anomaly detection model according to the train attribute information of the train body, wherein the train body anomaly detection model is constructed based on a deep learning sample detection algorithm; and inputting the preprocessed vehicle body side surface appearance image data and the preprocessed vehicle body bottom surface appearance image data into a vehicle body anomaly detection model, and outputting anomaly detection information by the vehicle body anomaly detection model. According to the method, the train body is subjected to anomaly detection by using the train body anomaly detection model constructed based on the deep learning sample detection algorithm, a detection model does not need to be constructed for each train body anomaly type, and the implementation difficulty of train body detection can be remarkably reduced.

Description

Technical Field

[0001] The present invention relates to the field of vehicle body detection technology, and in particular to a vehicle body anomaly detection method based on a deep learning small sample detection algorithm. Background Art

[0002] Train body anomaly detection is a critical area, impacting the safety and efficiency of train operations. This detection primarily involves capturing three-dimensional images of the train's sides and bottom to identify and detect various defects. Existing body anomaly detection is primarily based on defect-specific body anomaly recognition models, which typically require sufficient training to accurately identify specific defects. However, train body anomalies come in many different types and corresponding appearances. It would be unrealistic to pre-define a body anomaly recognition model for each anomaly type.

[0003] It can be seen that how to reduce the difficulty of implementing train body abnormality detection is a technical problem that needs to be solved urgently. Summary of the Invention

[0004] In this regard, the present invention provides a vehicle body anomaly detection method, system, electronic device, computer storage medium and computer program product based on a deep learning small sample detection algorithm to solve the above technical problems.

[0005] The present invention provides a vehicle body anomaly detection method based on a deep learning small sample detection algorithm, the method comprising the following steps:

[0006] Acquiring first appearance image data of a train body, the first appearance image data including body side appearance image data and body bottom appearance image data, and preprocessing the first appearance image data;

[0007] Retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm;

[0008] The pre-processed vehicle body side appearance image data and the vehicle body bottom appearance image data are respectively input into the vehicle body anomaly detection model, and the vehicle body anomaly detection model outputs anomaly detection information; wherein the anomaly detection information includes an abnormal area and an abnormality probability value.

[0009] Optionally, obtaining first appearance image data of a train body includes:

[0010] A high-speed line scan camera is used to perform 360-degree exterior imaging of the train to obtain first exterior image data, and the first exterior image data is divided into the vehicle body side exterior image data and the vehicle body bottom exterior image data.

[0011] Optionally, the preprocessing of the first appearance image data includes:

[0012] Retrieving side appearance image features and bottom appearance image features corresponding to the vehicle body anomaly detection model; wherein the side appearance image features and the bottom appearance image features are obtained based on training data of the vehicle body anomaly detection model;

[0013] The vehicle body side appearance image data is preprocessed according to the side appearance image features, and the vehicle body bottom appearance image data is preprocessed based on the bottom appearance image features.

[0014] Optionally, the side appearance image features and the bottom appearance image features obtained based on the training data of the vehicle body anomaly detection model are specifically:

[0015] Filtering the training data of the vehicle body anomaly detection model to obtain side surface group training data and bottom surface group training data to be put into use; wherein the side surface group training data includes a first training data set and a first test data set, and the bottom surface group training data includes a second training data set and a second test data set;

[0016] Using a convolutional network to perform feature extraction on the side group training data and the bottom group training data, respectively, to obtain side group image features and bottom group image features; wherein the side group image features include first training image features and first test image features, and the bottom group image features include second training image features and second test image features;

[0017] The first training image feature and the first test image feature, the second training image feature and the second test image feature are weightedly calculated to obtain the side appearance image feature and the bottom appearance image feature; wherein the weighted weight of the first training image feature is greater than that of the first test image feature, and the weighted weight of the second training image feature is greater than that of the second test image feature.

[0018] Optionally, the retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body includes:

[0019] Train speed interval data and travel geographical area data are parsed from the train attribute information of the train body, and the corresponding vehicle body abnormality detection model is retrieved based on the train speed interval data and the travel geographical area data.

[0020] Optionally, building the vehicle body anomaly detection model based on a deep learning small sample detection algorithm includes:

[0021] collecting a set of second appearance image data of a train body, and filtering out a portion of the set of second appearance image data to form small sample training data;

[0022] A general large model is selected, and the small sample training data is used to fine-tune the general large model until the training termination condition is met, thereby obtaining the vehicle body anomaly detection model.

[0023] The present invention also provides a vehicle body anomaly detection system based on a deep learning small sample detection algorithm, the system comprising an acquisition and preprocessing module, a model retrieval module, and an anomaly detection module;

[0024] The acquisition and preprocessing module is used to acquire first appearance image data of the train body, wherein the first appearance image data includes the side appearance image data and the bottom appearance image data of the train body, and preprocess the first appearance image data;

[0025] The model retrieval module is used to retrieve a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm;

[0026] The abnormality detection module is used to input the pre-processed vehicle side appearance image data and the vehicle bottom appearance image data into the vehicle body abnormality detection model respectively, and the vehicle body abnormality detection model outputs abnormality detection information; wherein, the abnormality detection information includes an abnormal area and an abnormality probability value.

[0027] The present invention also provides an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement any of the above methods.

[0028] The present invention also provides a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0029] The present invention also provides a computer program product, which implements any of the above methods when the computer program product is executed by a processor.

[0030] The present invention performs abnormality detection on train bodies by using a body anomaly detection model constructed based on a deep learning small sample detection algorithm. There is no need to build a detection model for each type of body anomaly, which can significantly reduce the difficulty of implementing train body detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without paying any creative work.

[0032] Figure 1 This is a flow chart of a vehicle body anomaly detection method based on a deep learning small sample detection algorithm disclosed in an embodiment of the present invention;

[0033] Figure 2 is a schematic diagram of a process for preprocessing first appearance image data disclosed in an embodiment of the present invention;

[0034] Figure 3 is a schematic diagram of a process for acquiring side appearance image features and bottom appearance image features disclosed in an embodiment of the present invention;

[0035] Figure 4 This is a structural diagram of a vehicle body anomaly detection system based on a deep learning small sample detection algorithm disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following specific embodiments illustrate the implementation of this application. Those familiar with the art can easily understand the other advantages and functions of this application from the content disclosed in this specification. Obviously, the described embodiments are part of the embodiments of this application, but not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0037] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0038] Existing body anomaly detection systems primarily rely on body anomaly recognition models for specific defects. These models typically require sufficient training to accurately identify specific defects. However, train body anomalies come in many different types and have corresponding appearances. It's clearly unrealistic to create a pre-defined body anomaly recognition model for each anomaly type.

[0039] Therefore, in order to reduce the difficulty of implementing train body anomaly detection, a technical problem that urgently needs to be solved, the present invention adopts the following technical means:

[0040] like Figure 1As shown, an embodiment of the present invention discloses a vehicle body anomaly detection method based on a deep learning small sample detection algorithm, the method comprising the following steps:

[0041] Acquiring first appearance image data of a train body, the first appearance image data including body side appearance image data and body bottom appearance image data, and preprocessing the first appearance image data;

[0042] Retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm;

[0043] The pre-processed vehicle body side appearance image data and the vehicle body bottom appearance image data are respectively input into the vehicle body anomaly detection model, and the vehicle body anomaly detection model outputs anomaly detection information; wherein the anomaly detection information includes an abnormal area and an abnormality probability value.

[0044] In response to the above technical problems, the present invention first obtains the side appearance image data and the bottom appearance image data of the train body respectively, and pre-processes them, including brightness adjustment, cropping, denoising, etc. In addition, the corresponding body anomaly detection model is retrieved according to the train attribute information of the train body, and the body anomaly detection model is pre-constructed using a deep learning sample detection algorithm. Finally, the pre-processed body side appearance image data and body bottom appearance image data obtained above are input into the body anomaly detection model respectively. After the body anomaly detection model performs an in-depth analysis on the first appearance image data, it can identify the area where the anomaly exists, and output the anomaly area and the anomaly probability value, guiding the inspection personnel to pay attention to the anomaly area, verify and determine the subsequent processing plan. It should be noted that the abnormal area can be the area where screws, cables, secondary protective equipment, pipelines, pantographs and other components are located.

[0045] Therefore, the present invention performs abnormality detection on the train body by using a body abnormality detection model constructed based on a deep learning sample detection algorithm. There is no need to build a detection model for each type of body abnormality, which can significantly reduce the difficulty of implementing train body detection.

[0046] Optionally, obtaining first appearance image data of a train body includes:

[0047] A high-speed line scan camera is used to perform 360-degree exterior imaging of the train to obtain first exterior image data, and the first exterior image data is divided into the vehicle body side exterior image data and the vehicle body bottom exterior image data.

[0048] In this embodiment, after the train to be tested enters the maintenance workshop, a high-speed line scan camera is used to perform 360° appearance imaging of the train, and the captured image data is then divided into vehicle side appearance image data and vehicle bottom appearance image data, which facilitates subsequent separate processing.

[0049] The present invention does not specifically limit the infrastructure for deploying the high-speed line scan camera, but it is required to include a camera angle adjustment mechanism. For example, the camera angle adjustment mechanism could be a curved chute surrounding the train body, within which the high-speed line scan camera can rotate using a sliding member, thereby achieving 360° imaging of the train body's exterior. Alternatively, the camera angle adjustment mechanism could be a multi-dimensionally adjustable robotic arm that carries the high-speed line scan camera to a position trajectory surrounding the train body, thereby achieving 360° imaging of the train body's exterior.

[0050] Alternatively, as Figure 2 As shown, the preprocessing of the first appearance image data includes:

[0051] Retrieving side appearance image features and bottom appearance image features corresponding to the vehicle body anomaly detection model; wherein the side appearance image features and the bottom appearance image features are obtained based on training data of the vehicle body anomaly detection model;

[0052] The vehicle body side appearance image data is preprocessed according to the side appearance image features, and the vehicle body bottom appearance image data is preprocessed based on the bottom appearance image features.

[0053] In this embodiment, in addition to conventional pre-processing such as brightness adjustment, cropping, and denoising, the first appearance image data also needs to be processed in a targeted manner to facilitate the recognition of the vehicle body anomaly detection model. To this end, the present invention is configured to retrieve the side appearance image features and bottom appearance image features corresponding to the vehicle body anomaly detection model, and the side appearance image features and bottom appearance image features are derived based on the training data used by the vehicle body anomaly detection model in the training phase. The side appearance image data and the bottom appearance image data of the vehicle body are then pre-processed based on the side appearance image features and the bottom appearance image features, respectively. In this way, the first appearance image data after pre-processing has a higher similarity with the training data used by the vehicle body anomaly detection model, thereby facilitating the vehicle body anomaly detection model to more accurately identify abnormal situations.

[0054] Alternatively, as Figure 3 As shown, the side appearance image features and the bottom appearance image features obtained based on the training data of the vehicle body abnormality detection model are specifically:

[0055] Filtering the training data of the vehicle body anomaly detection model to obtain side surface group training data and bottom surface group training data to be put into use; wherein the side surface group training data includes a first training data set and a first test data set, and the bottom surface group training data includes a second training data set and a second test data set;

[0056] Using a convolutional network to perform feature extraction on the side group training data and the bottom group training data, respectively, to obtain side group image features and bottom group image features; wherein the side group image features include first training image features and first test image features, and the bottom group image features include second training image features and second test image features;

[0057] The first training image feature and the first test image feature, the second training image feature and the second test image feature are weightedly calculated to obtain the side appearance image feature and the bottom appearance image feature; wherein the weighted weight of the first training image feature is greater than that of the first test image feature, and the weighted weight of the second training image feature is greater than that of the second test image feature.

[0058] In this embodiment, in order to improve the anomaly detection accuracy of the vehicle body anomaly detection model, the present invention performs directional preprocessing on the first appearance image data of the train body obtained by the high-speed line scan camera. That is, the first appearance image data is processed to be as close as possible to the training data used by the vehicle body anomaly detection model in the training phase (including the training data and test data actually put into use).

[0059] Specifically, the side and bottom training data sets that are actually put into use are first screened from the training data of the vehicle anomaly detection model. Both training data sets include both training data and test data. It should be noted that the vehicle anomaly detection model in the present invention is a composite model, which can be equipped with two sets of model structures, one for detecting anomalies on the side and the other for detecting anomalies on the bottom of the vehicle. Correspondingly, the two sets of training data are also used for separate training.

[0060] Next, a convolutional network is used to extract features from the side and bottom training data, respectively, to obtain side image features and bottom image features (both of which are the average of multiple side image features or bottom image features). The side image features represent the image features of each vehicle side data used to train the vehicle anomaly detection model, while the bottom image features represent the image features of each vehicle bottom data used to train the vehicle anomaly detection model. These image features include at least one of brightness, grayscale, clarity, saturation, contrast, and magnification.

[0061] Finally, a weighted calculation is performed on the first training image feature and the first test image feature in the side image feature group, and the second training image feature and the second test image feature in the bottom image feature group, respectively, to obtain the side appearance image feature and the bottom appearance image feature. The side appearance image feature represents the representative value of the image feature of each vehicle side data used to train the vehicle body anomaly detection model, and the bottom appearance image feature represents the representative value of the image feature of each vehicle bottom data used to train the vehicle body anomaly detection model.

[0062] It should be noted that since the vehicle abnormality detection model is mainly based on training data for model training, and the test data is only used to verify the stage training effect and adjust the training process of the model, the present invention sets the weight of the training image features corresponding to the training data to be greater than the weight of the test image features.

[0063] In addition, before preprocessing the side appearance image data of the vehicle body according to the obtained side appearance image features and preprocessing the bottom appearance image data of the vehicle body based on the bottom appearance image features, the side appearance image features and the bottom appearance image features can be restored to corresponding preprocessing types and indicators, such as image brightness value, contrast value, magnification, etc.

[0064] Optionally, the retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body includes:

[0065] Train speed interval data and travel geographical area data are parsed from the train attribute information of the train body, and the corresponding vehicle body abnormality detection model is retrieved based on the train speed interval data and the travel geographical area data.

[0066] In this embodiment, there are many types of trains, such as ordinary green-skinned trains, medium- and high-speed trains, EMUs, and high-speed trains. These trains vary significantly in speed and geographic area, each leading to different types of abnormalities in the train body. To address this, the present invention pre-classifies trains based on train speed range data and geographic area data, constructs corresponding body anomaly detection models for each type, and performs targeted training using corresponding appearance image data (with annotated information). Consequently, the corresponding body anomaly detection model can be derived based on the train speed range data and geographic area data.

[0067] Optionally, building the vehicle body anomaly detection model based on a deep learning small sample detection algorithm includes:

[0068] collecting a set of second appearance image data of a train body, and filtering out a portion of the set of second appearance image data to form small sample training data;

[0069] A general large model is selected, and the small sample training data is used to fine-tune the general large model until the training termination condition is met, thereby obtaining the vehicle body anomaly detection model.

[0070] In this embodiment, the vehicle anomaly detection model of the present invention is trained based on a general large model, such as the ChatGPT series. The vehicle anomaly detection model is obtained by fine-tuning the general large model. Fine-tuning training can be performed using methods such as fine-tuning, feature augmentation, and transfer learning. The details are explained as follows:

[0071] Fine-tuning: This is the most commonly used fine-tuning method. It adds a new classification layer to the last layer of the pre-trained model and then fine-tunes it based on the new dataset.

[0072] Feature augmentation: This method enhances the performance of the model by adding some artificial features to the data. These features can be manually designed or generated through automatic feature generation techniques.

[0073] Transfer learning: This approach uses a model trained on one task as the starting point for a new task, and then fine-tunes the model's parameters to adapt to the new task.

[0074] For fine-tuning training data, a group of second appearance image data of the train body is collected, and a part of the second appearance image data is filtered out from the group to constitute small sample training data. The small sample training data is used to perform targeted training on the general large model to obtain a body anomaly detection model suitable for body anomaly detection in the present invention.

[0075] like Figure 4 As shown, an embodiment of the present invention further discloses a vehicle body anomaly detection system based on a deep learning small sample detection algorithm, the system comprising an acquisition and preprocessing module, a model retrieval module, and an anomaly detection module;

[0076] The acquisition and preprocessing module is used to acquire first appearance image data of the train body, wherein the first appearance image data includes the side appearance image data and the bottom appearance image data of the train body, and preprocess the first appearance image data;

[0077] The model retrieval module is used to retrieve a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm;

[0078] The abnormality detection module is used to input the pre-processed vehicle side appearance image data and the vehicle bottom appearance image data into the vehicle body abnormality detection model respectively, and the vehicle body abnormality detection model outputs abnormality detection information; wherein, the abnormality detection information includes an abnormal area and an abnormality probability value.

[0079] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in the aforementioned embodiment.

[0080] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method described in the above embodiment.

[0081] The embodiment of the present invention further discloses a computer program product, which implements the method described in the above embodiment when the computer program product is executed by a processor.

[0082] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0083] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0084] In the context of the present disclosure, computer storage media can be tangible media that can contain or store programs for use by or in combination with an instruction execution system, device or equipment. Machine-readable media can be machine-readable signal media or machine-readable storage media. Computer storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of computer storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0085] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0086] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), the Internet, and a blockchain network.

[0087] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0088] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0089] Although the embodiments or examples of the present disclosure have been described with reference to the accompanying drawings, it should be understood that the above-mentioned methods, systems and devices are merely exemplary embodiments or examples, and the scope of the present disclosure is not limited by these embodiments or examples, but is only limited by the claims after authorization and their equivalents. Various elements in the embodiments or examples can be omitted or replaced by their equivalents. In addition, the steps can be performed in an order different from that described in the present disclosure. Further, the various elements in the embodiments or examples can be combined in various ways. It is important that as technology evolves, many of the elements described here can be replaced by equivalent elements that appear after the present disclosure.

Claims

1. A vehicle body anomaly detection method based on a deep learning small sample detection algorithm, characterized by: The method comprises the following steps: Acquiring first appearance image data of a train body, the first appearance image data including body side appearance image data and body bottom appearance image data, and preprocessing the first appearance image data; Retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm; The pre-processed vehicle body side appearance image data and the vehicle body bottom appearance image data are respectively input into the vehicle body anomaly detection model, and the vehicle body anomaly detection model outputs anomaly detection information; wherein the anomaly detection information includes an abnormal area and an abnormality probability value.

2. The vehicle body anomaly detection method based on a deep learning small sample detection algorithm according to claim 1, characterized in that: Acquiring first appearance image data of a train body, including: A high-speed line scan camera is used to perform 360-degree exterior imaging of the train to obtain first exterior image data, and the first exterior image data is divided into the vehicle body side exterior image data and the vehicle body bottom exterior image data.

3. The vehicle body anomaly detection method based on a deep learning small sample detection algorithm according to claim 2, characterized in that: Preprocessing the first appearance image data includes: Retrieving side appearance image features and bottom appearance image features corresponding to the vehicle body anomaly detection model; wherein the side appearance image features and the bottom appearance image features are obtained based on training data of the vehicle body anomaly detection model; The vehicle body side appearance image data is preprocessed according to the side appearance image features, and the vehicle body bottom appearance image data is preprocessed based on the bottom appearance image features.

4. The vehicle body anomaly detection method based on a deep learning small sample detection algorithm according to claim 3 is characterized by: The side appearance image features and the bottom appearance image features obtained based on the training data of the vehicle body anomaly detection model are specifically: Filtering the training data of the vehicle body anomaly detection model to obtain side surface group training data and bottom surface group training data to be put into use; wherein the side surface group training data includes a first training data set and a first test data set, and the bottom surface group training data includes a second training data set and a second test data set; Using a convolutional network to perform feature extraction on the side group training data and the bottom group training data, respectively, to obtain side group image features and bottom group image features; wherein the side group image features include first training image features and first test image features, and the bottom group image features include second training image features and second test image features; The first training image feature and the first test image feature, the second training image feature and the second test image feature are weightedly calculated to obtain the side appearance image feature and the bottom appearance image feature; wherein the weighted weight of the first training image feature is greater than that of the first test image feature, and the weighted weight of the second training image feature is greater than that of the second test image feature.

5. The vehicle body anomaly detection method based on a deep learning small sample detection algorithm according to claim 4 is characterized in that: Retrieving a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, including: Train speed interval data and travel geographical area data are parsed from the train attribute information of the train body, and the corresponding vehicle body abnormality detection model is retrieved based on the train speed interval data and the travel geographical area data.

6. The vehicle body anomaly detection method based on a deep learning small sample detection algorithm according to claim 1, characterized in that: The vehicle body anomaly detection model is constructed based on the deep learning sample detection algorithm, including: collecting a set of second appearance image data of a train body, and filtering out a portion of the set of second appearance image data to form small sample training data; A general large model is selected, and the small sample training data is used to fine-tune the general large model until the training termination condition is met, thereby obtaining the vehicle body anomaly detection model.

7. A vehicle anomaly detection system based on a deep learning small sample detection algorithm, comprising an acquisition and preprocessing module, a model retrieval module, and an anomaly detection module; characterized in that: The acquisition and preprocessing module is used to acquire first appearance image data of the train body, wherein the first appearance image data includes the side appearance image data and the bottom appearance image data of the train body, and preprocess the first appearance image data; The model retrieval module is used to retrieve a corresponding vehicle body anomaly detection model according to the train attribute information of the train body, wherein the vehicle body anomaly detection model is constructed based on a deep learning sample detection algorithm; The abnormality detection module is used to input the pre-processed vehicle side appearance image data and the vehicle bottom appearance image data into the vehicle body abnormality detection model respectively, and the vehicle body abnormality detection model outputs abnormality detection information; wherein, the abnormality detection information includes an abnormal area and an abnormality probability value.

8. An electronic device comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 6.

9. A computer storage medium storing a computer program, wherein: The computer program is executed by a processor to implement the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that: When the computer program product is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images