Abnormality detection method, device, equipment and storage medium for new energy vehicle PCB board

Abstract

The present application provides a method, device, equipment and storage medium for detecting abnormalities of a PCB board of a new energy vehicle. The method comprises: collecting a PCB image and at least one supplementary image of a PCB board of a new energy vehicle; extracting features of the PCB image and the supplementary image using a multimodal fusion network; performing feature reconstruction processing on multimodal image features to obtain reconstructed multimodal image features; performing attention feature extraction on the reconstructed multimodal image features to obtain PCB board image features of wiring density and component integration of the PCB board of the new energy vehicle; performing feature distance calculation processing on PCB board image features and PCB board sample image features to obtain a feature distance image of the PCB board of the new energy vehicle; detecting abnormal areas of the PCB board of the new energy vehicle based on the feature distance image of the PCB board of the new energy vehicle and outputting location information of the abnormal areas; the scheme can integrate multimodal image data of the PCB board of the new energy vehicle to improve the accuracy and efficiency of detection.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and storage medium for detecting anomalies of a PCB board of a new energy vehicle. Background Art

[0002] With the rapid development of the new energy vehicle industry, higher requirements are placed on the electronic control system of vehicles. As a key component in electronic equipment, the quality and reliability of printed circuit boards (PCBs) directly affect the performance and safety of new energy vehicles. Therefore, efficient and accurate abnormality detection of PCBs has become particularly important.

[0003] Traditional PCB inspection methods mainly rely on manual visual inspection or simple automated inspection equipment. These methods have the following limitations: First, manual inspection is inefficient and easily affected by subjective factors, making it difficult to ensure consistency and accuracy; second, simple automated inspection equipment can often only detect obvious defects and has difficulty identifying minor or complex anomalies. Summary of the invention

[0004] The embodiments of the present application provide a method, device, equipment and storage medium for detecting anomalies of a PCB board of a new energy vehicle, which can improve the accuracy of detecting anomalies of a PCB board of a new energy vehicle.

[0005] One aspect of the present application provides a method for detecting abnormalities of a PCB board of a new energy vehicle, comprising:

[0006] Collecting a PCB image and at least one supplementary image of a PCB board of a new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image;

[0007] Using a multimodal fusion network to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board;

[0008] Performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features;

[0009] Performing attention feature extraction on the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board;

[0010] Performing feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, wherein the PCB board sample image is an image feature of a sample PCB board based on design parameters of the new energy vehicle PCB board, and the design parameters are parameters focusing on standard wiring density and component integration of the new energy vehicle PCB board;

[0011] Based on the characteristic distance image of the new energy vehicle PCB board, an abnormal area of ​​the new energy vehicle PCB board is detected and the position information of the abnormal area is output.

[0012] Another aspect of the present application provides an abnormality detection device for a new energy vehicle PCB board, comprising:

[0013] An acquisition module, used to acquire a PCB image and at least one supplementary image of a PCB board of a new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image;

[0014] A feature extraction module, used to extract features from the PCB image and the supplementary image using a multimodal fusion network to obtain multimodal image features of the new energy vehicle PCB board;

[0015] A feature reconstruction module, used for performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features;

[0016] An attention feature module is used to extract attention features from the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board;

[0017] A distance acquisition module is used to perform feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, wherein the PCB board sample image is an image feature of a sample PCB board based on the design parameters of the new energy vehicle PCB board, and the design parameters are parameters focusing on the standard wiring density and component integration of the new energy vehicle PCB board;

[0018] The detection module is used to detect the abnormal area of ​​the new energy vehicle PCB board based on the characteristic distance image of the new energy vehicle PCB board and output the position information of the abnormal area.

[0019] Another aspect of the present application provides a computer device, comprising: a memory and a processor; wherein the memory is used to store a program;

[0020] The processor is used to execute the program in the memory, including executing the methods in the above aspects.

[0021] Another aspect of the present application provides a computer-readable storage medium, in which instructions are stored. When the computer-readable storage medium is run on a computer, the computer is enabled to execute the above-mentioned methods.

[0022] Another aspect of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the methods provided by the above aspects.

[0023] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0024] The present application provides an abnormality detection and device for a new energy vehicle PCB board, and the method includes: collecting a PCB image and at least one supplementary image of a new energy vehicle PCB board, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image; using a multimodal fusion network to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board; performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features; performing attention feature extraction on the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board; performing feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, wherein the PCB board sample image is an image feature of a sample PCB board based on the design parameters of the new energy vehicle PCB board, and the design parameters are parameters that focus on the standard wiring density and component integration of the new energy vehicle PCB board; detecting an abnormal area of ​​the new energy vehicle PCB board based on the new energy vehicle PCB board feature distance image and outputting location information of the abnormal area. This solution can integrate multimodal image data of new energy vehicle PCB boards, which can improve the accuracy and efficiency of detection; and this solution can adapt to the special wiring density and component integration of new energy vehicle PCBs, and can further identify and locate small defects and potential anomalies on new energy vehicle PCBs, thereby improving the reliability and safety of new energy vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of a PCB board provided in a certain embodiment of the present application;

[0026] Figure 2 A schematic diagram of the architecture of a PCB board abnormality detection system provided in a certain embodiment of the present application;

[0027] Figure 3 A flowchart of a PCB board abnormality detection method provided in an embodiment of the present application;

[0028] Figure 4 A schematic diagram of abnormal identification of a PCB board provided in a certain embodiment of the present application;

[0029] Figure 5 A schematic diagram of the structure of an abnormality detection device for a new energy vehicle PCB board provided in a certain embodiment of the present application;

[0030] Figure 6 A schematic diagram of a server structure provided for a certain embodiment of the present application. DETAILED DESCRIPTION

[0031] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the implementation of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "corresponding to" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] The field of artificial intelligence (AI) is dedicated to simulating and expanding human intelligence, involving theories, methods, technologies and application systems for perceiving the environment, acquiring knowledge and applying knowledge to achieve the best results. As a branch of computer science, AI aims to understand the nature of intelligence and create machines that can simulate human intelligent responses, with perception, reasoning and decision-making capabilities.

[0033] AI technology is an interdisciplinary field that covers both hardware and software technologies. Hardware includes sensors, AI chips, etc., while software includes computer vision, speech processing, natural language processing, and machine learning / deep learning. Among them, computer vision (CV) technology enables machines to "see", that is, to use cameras and computers instead of human eyes to perform recognition, measurement, and image processing to obtain information from images or multi-dimensional data. Computer vision technology includes image processing, recognition, semantic understanding, retrieval, OCR, video processing, semantic understanding, content / behavior recognition, three-dimensional reconstruction, 3D technology, virtual reality, augmented reality, simultaneous positioning and map construction, as well as biometric recognition technologies such as face recognition and fingerprint recognition.

[0034] Printed circuit boards (PCBs) are core components of modern electronic devices, and their quality directly affects product performance. PCB defect detection is a key link in the production line and is crucial to the performance and safety of electronic products. Common PCB defects include missing solder holes and missing copper wires. Figure 1 In the new energy vehicle industry, printed circuit boards (PCBs) are key components in electronic devices, and their quality and reliability directly affect the performance and safety of new energy vehicles. Therefore, efficient and accurate abnormality detection of PCBs has become particularly important. With the development of electronic technology, the design and manufacture of PCBs have become more complex, and traditional detection methods can no longer meet the detection needs of modern PCBs. Especially in the field of new energy vehicles, PCBs must not only withstand high voltages and high currents, but also maintain stable operation under extreme temperatures and vibration environments, which places higher requirements on the wiring density and component integration of PCBs.

[0035] Traditional PCB inspection methods for new energy vehicles rely on manual visual inspection, which has problems such as subjective errors, slow inspection speed, high cost and difficulty in data collection. With the development of PCB technology, the requirements for inspection technology have become more stringent, and traditional visual inspection can no longer meet the needs of modern production. The development of machine vision inspection technology enables PCB abnormalities to be discovered in a timely manner, improving the efficiency and accuracy of inspection.

[0036] The abnormality detection method of the new energy vehicle PCB board provided in the embodiment of the present application uses a multi-modal fusion network to extract and fuse the PCB board images of multiple different modes, which can integrate the multi-modal image data of the new energy vehicle PCB board, and detect the abnormal area of ​​the PCB board based on the features of the multi-modal image, which can improve the accuracy and efficiency of the detection. The embodiment of the present application also takes into account the specific requirements of the new energy vehicle PCB for wiring density and component integration, extracts related features from the multi-modal image features, and performs abnormality detection of the PCB board based on the features. It can be seen that the scheme can adapt to the special wiring density and component integration of the new energy vehicle PCB, and can further identify and locate the small abnormal defects and potential abnormalities on the new energy vehicle PCB under the wiring density and component integration requirements, further improve the accuracy of abnormality detection of the new energy vehicle PCB board, and thus improve the reliability and safety of the new energy vehicle.

[0037] In one embodiment, for ease of understanding, please refer to Figure 2 , Figure 2 This is an application environment diagram of abnormality detection of a new energy vehicle PCB board in an embodiment of the present application, such as Figure 2 As shown, the abnormality detection method of the PCB board of a new energy vehicle in the embodiment of the present application is applied to an image-based PCB board abnormality detection system. The PCB board abnormality detection system includes: a server, a multimodal image acquisition device and a terminal device; wherein the server can be an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms. The terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, etc., but is not limited to this. The terminal and the server can be directly or indirectly connected by wired or wireless communication, and the embodiment of the present application is not limited here.

[0038] Among them, the multimodal image acquisition device may include a visible light camera such as an ordinary camera, and an invisible light image acquisition device such as an infrared camera, an X-ray image acquisition device, and the like.

[0039] The server collects a PCB image and at least one supplementary image of a new energy vehicle PCB board, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image; a multimodal fusion network is used to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board; feature reconstruction is performed on the multimodal image features to obtain reconstructed multimodal image features; attention feature extraction is performed on the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board; feature distance calculation is performed on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, the PCB board sample image is an image feature of a sample PCB board based on the design parameters of the new energy vehicle PCB board, and the design parameters are parameters that focus on the standard wiring density and component integration of the new energy vehicle PCB board; based on the new energy vehicle PCB board feature distance image, the abnormal area of ​​the new energy vehicle PCB board is detected and the location information of the abnormal area is output. In one embodiment, the abnormal area information of the new energy vehicle PCB board is also sent to the terminal, such as sending the PCB image and marking the abnormal area in the image. In addition, the abnormal area information also includes the abnormal type of the abnormal area, such as leaking holes, excess copper, holes, burrs, gaps, hole position offsets, insufficient etching and incorrect punching.

[0040] The following is an introduction to the abnormality detection method of the new energy vehicle PCB board in this application from the perspective of the server. Figure 3 The anomaly detection method provided in the embodiment of the present application is specifically:

[0041] S110, collecting a PCB image and at least one supplementary image of a PCB board of a new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image.

[0042] In the embodiments of the present application, visible light images refer to images of printed circuit boards (PCBs) captured using standard optical photography techniques, which reflect the PCB features within the spectrum visible to the human eye. Visible light images can show the appearance features of the PCB, including but not limited to circuit paths, component layouts, and external defects.

[0043] Supplementary images are images that provide additional information in addition to visible light images. These images are captured using different physical principles and can reveal non-visible features of PCBs. Supplementary images include at least:

[0044] Infrared images, which are captured by detecting and recording infrared radiation emitted or reflected by the PCB, can reveal the temperature distribution of the PCB, thermal defects, and thermal performance of the components;

[0045] X-ray images, which use the ability of X-rays to penetrate materials to capture the internal structure of the PCB, including internal layers, hidden connections, and potential internal defects.

[0046] S120, using a multimodal fusion network to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board.

[0047] In the embodiments of the present application, multimodal image features refer to a set of features extracted from image data of different modalities, which integrates the unique information provided by various image modalities. Specifically, in one embodiment, multimodal image features specifically refer to the integration of features extracted from PCB board images of different modalities such as visible light images, infrared images, and X-ray images.

[0048] Specifically, visible light image features include but are not limited to visual features such as the geometric shape, texture, color, and layout of the PCB board; infrared image features include thermal features such as the temperature distribution, thermal anomalies, and thermal conductivity characteristics of the PCB board; X-ray image features include internal features such as the internal structure of the PCB board, hidden traces, and solder joints. The specific feature set can be set according to actual needs.

[0049] Among them, the embodiment of the present application can use a multimedia fusion network for multimodal image features, which can improve the accuracy and efficiency of image features. In one embodiment, the multimodal fusion network includes: multiple input branch networks, a feature fusion layer, and a deep learning core layer, wherein each input branch corresponds to an image feature extraction of a modality;

[0050] The multimodal fusion network is used to extract features from the PCB image and the supplementary image to obtain the multimodal image features of the new energy vehicle PCB board, including:

[0051] The image corresponding to each modality is input into the branch network for feature extraction to obtain image features of different modalities;

[0052] A feature fusion layer is used to map the image features corresponding to different modalities into a common feature space, and the mapped features are fused in the common feature space to obtain fused features;

[0053] The deep learning core layer is used to perform deep feature extraction on the fused features and the multimodal image features of the PCB board of new energy vehicles.

[0054] Specifically, the structure of the multimodal fusion network is as follows:

[0055] Multiple input branches, each branch corresponds to an image of a modality, specifically including: i. a visible light image branch, used to receive and process a visible light image of a PCB board; ii. at least one supplementary image branch, used to receive and process a supplementary image of a different modality from the visible light image, such as an infrared image branch and an X-ray image branch;

[0056] The feature fusion layer is used to fuse features from different modal branches, including: i. A feature mapping unit that maps the deep features extracted from each modal branch into a common feature space; ii. A fusion strategy unit that uses weighted fusion, concatenation fusion, or attention mechanism to effectively integrate multimodal features;

[0057] The deep learning core layer is used for deep feature extraction based on fused features, including: i. Multiple convolutional layers for extracting multi-level features of the image; ii. Multiple pooling layers for reducing the spatial dimension of the features and extracting abstract features of the image; iii. Fully connected layers for mapping the extracted deep features to the final feature representation; iv. Activation functions, such as ReLU or Sigmoid functions, for introducing nonlinearity and enhancing the expressiveness of the network; v. Normalization layers, such as Batch Normalization, for stabilizing the network training process and accelerating convergence.

[0058] The feature fusion layer also includes a gating mechanism for dynamically adjusting the weights of different modal features in the fusion process to highlight features that are more sensitive to anomaly detection; in one embodiment, the deep learning core layer also includes a feature enhancement module for enhancing features in the fused features that are related to anomaly detection of new energy vehicle PCB boards and suppressing irrelevant or noise features.

[0059] The multimodal fusion network is trained in an end-to-end manner to optimize the network parameters so that the extracted multimodal image features have higher sensitivity and accuracy for anomaly detection of new energy vehicle PCB boards.

[0060] The multimodal fusion network training method includes key steps such as data preparation, network architecture design, loss function definition, end-to-end training, performance evaluation, and hyperparameter adjustment. In the data preparation step, multimodal PCB board image data, including visible light images, infrared images, and X-ray images, are collected and preprocessed to meet the requirements of network input. Subsequently, in the network architecture design step, a multimodal fusion network with multiple input branches is constructed, each branch specializing in processing image data of one modality, and including a feature extraction layer, a feature fusion layer, and a feature output layer. Next, in the loss function definition step, a loss function is determined to measure the difference between the network output and the actual anomaly label, and a suitable loss function is selected to meet the needs of the anomaly detection task.

[0061] In the end-to-end training step, the network is iteratively optimized from the input layer to the output layer, and the loss function is minimized using the back-propagation algorithm and the gradient descent method to optimize the network parameters. In the performance evaluation step, the network performance is evaluated using indicators such as accuracy, recall, and F1 score on an independent validation set to ensure the sensitivity and accuracy of anomaly detection. Based on the evaluation results, the learning rate, batch size, etc. are adjusted in the hyperparameter adjustment step to further improve the network performance. Finally, in the model fine-tuning step, the trained network model is fine-tuned to adapt it to a specific type of PCB board or anomaly type. Finally, in the deployment and application step, the trained multimodal fusion network is deployed to the actual PCB board anomaly detection system to achieve real-time or batch anomaly detection.

[0062] S130 , performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features.

[0063] In one embodiment, in order to improve the accuracy of features and extract features related to wiring density and component integration characteristics, the multimodal feature image can be compressed and reconstructed to enhance the representation of abnormal features, especially for the wiring density and component integration characteristics of new energy vehicle PCBs, as well as potential abnormal patterns related to these characteristics.

[0064] Specifically, in order to improve the efficiency and accuracy of feature extraction, in one embodiment, the multimodal image features are subjected to dimensionality reduction processing; according to the wiring density and component integration of the PCB board of the new energy vehicle, the multimodal image features after dimensionality reduction are subjected to the use of a deep convolutional autoencoder network to remove redundant information and retain key features; the compressed features are upsampled and reconstructed through the decoder part of the deep convolutional autoencoder network to obtain the reconstructed multimodal image features.

[0065] Among them, the deep convolutional autoencoder network includes multiple encoding layers, each encoding layer reduces the spatial dimension of the features through convolution operations and pooling operations, and extracts a more compact feature representation; the decoder part includes multiple decoding layers, each decoding layer gradually restores the spatial resolution of the features through deconvolution operations or transposed convolution operations, and enhances the representation of the wiring density and component integration characteristics of the new energy vehicle PCB board.

[0066] In one embodiment, a deep convolutional autoencoder network can be pre-trained to reconstruct features. Specifically, the training process begins by feeding the input data (such as a sample multimodal image of a PCB board) into the encoder part of the autoencoder. The encoder part consists of multiple convolutional layers, each of which extracts the features of the input data through a filter, and gradually reduces the spatial dimension of the data while increasing the depth of the features. Then, the output of the encoder (i.e., the low-dimensional feature representation) is fed into the decoder part. The structure of the decoder part is opposite to that of the encoder, and the spatial dimension of the data is gradually restored through a series of deconvolution layers, and an output similar to the input data is finally reconstructed. During the training process, the network weights are adjusted by minimizing the difference between the input and the reconstructed output (for example, using the mean square error as the loss function), and this process usually uses a gradient descent algorithm. In this way, the data representation learned by the autoencoder can not only capture the important features of the input data, but also perform effective data compression and denoising without the need for labels, thereby realizing the training of the deep convolutional autoencoder network.

[0067] S140, performing attention feature extraction on the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board.

[0068] In the embodiments of this application, wiring density refers to the distribution frequency and complexity of the conductive tracks (routings) on the PCB board of new energy vehicles, which is determined according to the requirements of circuit design and the standards of electrical performance. Wiring density involves the width, spacing and inter-layer layout of the routing, which together determine the electrical performance of the PCB board, such as signal transmission speed, anti-interference ability and thermal management.

[0069] Component integration refers to the density of electronic components on a PCB per unit area, including the size, quantity, and arrangement of components. Highly integrated PCB design can improve space utilization and reduce the size of the circuit board, but it also places higher requirements on heat dissipation, signal integrity, and electromagnetic compatibility.

[0070] For new energy vehicles, the wiring density and component integration of PCB boards are particularly critical because they directly affect the performance and reliability of the vehicle's electronic control system. High-density wiring and highly integrated components can provide more functions and higher processing capabilities, but also require more sophisticated manufacturing processes and stricter quality control. In new energy vehicles, PCB boards often need to work stably under harsh environments such as high temperature, vibration and electromagnetic interference. Therefore, the method in the present invention is particularly focused on detecting PCB boards that do not meet the standard wiring density and component integration requirements to ensure that they can operate reliably in new energy vehicles for a long time.

[0071] An attention mechanism is applied to the reconstructed multimodal feature image to further refine and highlight these features by focusing on the feature maps of wiring density and component integration, where the attention mechanism can be a channel attention or spatial attention based strategy to enhance the feature representation most relevant to anomaly detection of wiring density and component integration in new energy vehicle PCB boards.

[0072] The reconstructed multimodal image features are preprocessed to ensure data consistency and applicability. This step may include normalization, scaling, and removal of irrelevant features to highlight the key features of new energy vehicle PCB boards, especially wiring density and component integration. Next, an attention mechanism model is applied that can identify and focus on features related to wiring density and component integration. The attention mechanism can be channel-based or spatial-based attention, or a combination of both to ensure that features are captured from different angles.

[0073] When performing attention feature extraction, the model will automatically learn to assign different weights to different feature maps. These weights reflect the importance of the feature maps to the final task - identifying the wiring density and component integration of the PCB board. In this way, the model is able to focus resources and computing power on the most relevant features, thereby improving the efficiency and accuracy of feature extraction. This process involves training a deep learning network that can automatically adjust weights based on a large amount of labeled data to maximize the performance of feature extraction.

[0074] After the attention feature extraction is completed, the obtained PCB board image features will be used for further analysis and processing. These features will be input into a classifier or regression model to predict whether the PCB board has defects or evaluate its quality grade. In order to optimize the performance of the model, some post-processing steps may be required, such as feature selection, dimensionality reduction, or model fusion. These steps help reduce redundant information, improve the generalization ability of the model, and ultimately improve the accuracy of anomaly detection.

[0075] S150, performing feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, where the PCB board sample image is an image feature of a sample PCB board based on design parameters of the new energy vehicle PCB board.

[0076] Among them, the design parameters focus on the standard wiring density and component integration of new energy vehicle PCB boards.

[0077] Among them, the new energy vehicle PCB feature distance image refers to an image obtained by calculating the distance between the image features of the new energy vehicle PCB and a set of predefined sample PCB image features. This image intuitively represents the degree of deviation between the PCB features and the standard design parameters, wherein the distance calculation can be based on a variety of measurement methods, including but not limited to Euclidean distance, Manhattan distance or cosine similarity.

[0078] In one embodiment, the PCB board image features and the PCB board sample image features can be converted into feature vectors, and the distance between the extracted feature vectors can be calculated. Various distance metrics can be used in this step, such as Euclidean distance, Manhattan distance or cosine similarity. The choice of distance metric depends on the type of feature and the specific requirements of the problem. Finally, a feature distance image is generated based on the calculated feature distance. This image maps the feature distance to the geometric layout of the PCB board, and is usually represented by a heat map or other visualization technology. In the heat map, different colors represent different distance values, so that the area on the PCB board that is most different from the sample image can be intuitively identified.

[0079] S160, detecting an abnormal area of ​​the new energy vehicle PCB board based on the new energy vehicle PCB board feature distance image and outputting position information of the abnormal area.

[0080] After obtaining the characteristic distance image of the new energy vehicle PCB board, the server can detect the abnormal area of ​​the new energy vehicle PCB board according to the image and output the location information of the abnormal area. In one embodiment, a distance threshold characterizing the abnormality of a pixel unit is obtained; the characteristic distance between each pixel unit in the PCB board image and the corresponding pixel unit in the PCB board sample image is compared with the distance threshold, and the abnormal result of each pixel unit in the PCB board image is determined according to the comparison result; the abnormal area is determined in the PCB board image according to the abnormal result of each pixel unit, and the location information of the abnormal area is output.

[0081] Specifically, in order to improve the accuracy of abnormal area detection, the following steps can be adopted:

[0082] Determine one or more thresholds based on historical data and statistical analysis to distinguish normal and abnormal feature distances, where the thresholds can be fixed or dynamically adjusted and optimized according to the specific type and production batch of the PCB board;

[0083] In the abnormal region detection step, the feature distance image is compared with the threshold value, and the feature distance regions exceeding the threshold value are identified, which are considered as potential abnormal regions;

[0084] The abnormal region verification step further analyzes the initially identified abnormal region to exclude false positives introduced during image acquisition or processing, such as verifying the authenticity of the abnormal region through morphological operations or machine learning classifiers;

[0085] The abnormal area positioning step is to accurately locate the confirmed abnormal area, determine the center coordinates or boundary box of the abnormal area, and record these location information;

[0086] The abnormal area output step outputs the location information of the abnormal area in the form of image annotation or data output, providing the specific location, size and possible abnormal type of the abnormal area for subsequent inspection and processing. Figure 4 ,Through the above method, the abnormal area of ​​the PCB board of the new energy vehicle can be detected and ,marked with a box in the PCB board image.

[0087] In order to improve the accuracy of anomaly detection, if the new energy vehicle PCB board feature distance image includes pixel features, step S160 may specifically include:

[0088] Extract the pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image;

[0089] The pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image are processed to obtain abnormal probability information of each pixel unit in the PCB image;

[0090] Based on the abnormal probability information of each pixel unit in the PCB image, the abnormal area of ​​the new energy vehicle PCB board is detected and the position information of the abnormal area is output.

[0091] Specifically, the pixel features corresponding to each pixel unit are extracted from the image of the new energy vehicle PCB board. These features may include information such as color, brightness, texture and shape. This step is achieved through image processing technology, such as local binary pattern (LBP), Gabor filter or deep learning feature extraction network. At the same time, we also extract the pixel features corresponding to each pixel unit from the feature distance image. These features represent the distance between the PCB board features and the sample features, thereby reflecting the deviation between the PCB board features and the design parameters.

[0092] Then, the extracted pixel features are compared and analyzed. Using machine learning algorithms such as neural networks, the features of each pixel unit are classified and evaluated, and its abnormal probability is calculated. In this process, we consider the spatial relationship and contextual information between pixel features to improve the accuracy of anomaly detection. For example, image segmentation technology can be used to identify and analyze components and wiring patterns on PCB boards. According to the abnormal probability information of each pixel unit, a threshold is set to distinguish normal and abnormal pixel units. Pixel units with abnormal probabilities higher than the threshold are considered as potential abnormal areas. The spatial distribution of these abnormal pixel units on the PCB board is further analyzed, and clustering algorithms such as K-means or DBSCAN are used to identify continuous abnormal areas. Finally, the location information of the abnormal area is determined, including the center coordinates, bounding box or area of ​​the area, and this information is recorded for subsequent analysis and processing. The location information of the abnormal area is output in a visual way (such as marking the abnormal area on the original PCB board image) or in a structured data format (such as JSON or XML file).

[0093] In order to improve the accuracy and efficiency of anomaly detection, the pixel feature corresponding to each pixel unit in the PCB board image feature can be set as the key, and the pixel feature corresponding to each pixel unit in the new energy vehicle PCB board feature distance image can be set as the value. The pixels corresponding to the pixel units in the new energy vehicle PCB board feature distance image are subjected to feature processing of the mutual attention mechanism to obtain the final pixel feature; the abnormal probability information of each pixel unit in the PCB board image is calculated according to the final pixel feature.

[0094] The mutual attention mechanism is used to calculate the similarity between the pixel unit features of each pixel unit in the PCB board image features and the pixel unit features of the pixel unit in the feature difference reconstructed image, and the abnormal probability information corresponding to each pixel unit is obtained, thereby improving the accuracy of the calculation of the abnormal probability information corresponding to the pixel unit.

[0095] Specific:

[0096] The pixel feature corresponding to each pixel unit in the PCB board image feature is set as the key, and the pixel feature corresponding to each pixel unit in the new energy vehicle PCB board feature distance image is set as the value. The pixels corresponding to the pixel units in the new energy vehicle PCB board feature distance image are subjected to the mutual attention mechanism to obtain the final pixel feature.

[0097] It can be understood that if the PCB board image feature (which can be presented in the form of an image) is F, and the pixel feature of the PCB board feature distance image is D, for each pixel point i1 on F, the pixel point feature corresponding to each pixel point i1 is used as the key, and the pixel point i2 on D and the P pixels corresponding to the pixel point i2 are used as the value (where the pixel point i1 on F corresponds to the pixel point i2 on D), the beneficial information in the feature distance image is dynamically extracted through the mutual attention mechanism to obtain the first pixel point feature. The first pixel point feature can be expressed by the following formula:

[0098]

[0099] Among them, q (query) and k (key) are the pixel features corresponding to each pixel in the PCB board image feature (that is, the pixel features corresponding to the pixels in the PCB board image feature are used as the key attribute feature k and the query feature q respectively), v is the pixel feature corresponding to each pixel in the new energy vehicle PCB board feature distance image, dk is the dimension or number of channels of the input feature, K T The transpose of the pixel features corresponding to each pixel (i.e. the transpose of the key).

[0100] After obtaining the final pixel features of the PCB board image, a classification function such as the Softmax function can be used to generate the abnormal probability. The feature vector of the final pixel feature is input into the Softmax function, which maps the feature vector into a probability distribution so that the probability value corresponding to each feature vector is between 0 and 1, and the sum of all elements is 1, thereby obtaining the abnormal probability of the pixel unit (here, the abnormal probability of the pixel unit corresponding to the feature vector with a distribution probability less than a certain value can be set to be higher).

[0101] In one embodiment, the accuracy of PCB board abnormality detection is improved. After obtaining the abnormal probability information of each pixel unit in the PCB image, the pixel units that may be abnormal (such as the pixel units corresponding to the feature vectors with a distribution probability less than 1%) can be initially screened and selected according to the probability information, and then these pixel units are further verified. Specifically, the candidate abnormal pixel units in the PCB image are determined based on the abnormal probability information of each pixel unit in the PCB image;

[0102] Obtaining a distance threshold representing an abnormality of a pixel unit;

[0103] Compare the characteristic distance between each candidate abnormal pixel unit in the PCB board image and the corresponding pixel unit in the PCB board sample image with the distance threshold, and determine the abnormal result of each candidate abnormal pixel unit in the PCB board image according to the comparison result;

[0104] An abnormal area is determined in the PCB board image according to the abnormal result of each candidate abnormal pixel unit, and position information of the abnormal area is output.

[0105] For example, after obtaining the abnormal probability information of each pixel unit, the candidate pixel unit, i.e., the pixel unit that may be abnormal, is determined according to the probability;

[0106] Determine one or more thresholds based on historical data and statistical analysis to distinguish normal and abnormal feature distances, where the thresholds can be fixed or dynamically adjusted and optimized according to the specific type and production batch of the PCB board;

[0107] The abnormal area detection step is to compare the characteristic distance between each candidate abnormal pixel unit in the PCB board image and the corresponding pixel unit in the PCB board sample image, identify the candidate abnormal pixel units exceeding the threshold, and construct the abnormal area;

[0108] The abnormal region verification step further analyzes the initially identified abnormal region to exclude false positives introduced during image acquisition or processing, such as verifying the authenticity of the abnormal region through morphological operations or machine learning classifiers;

[0109] The abnormal area positioning step is to accurately locate the confirmed abnormal area, determine the center coordinates or boundary box of the abnormal area, and record these location information;

[0110] The abnormal area output step outputs the location information of the abnormal area in the form of image annotation or data output, providing the specific location, size and possible abnormal type of the abnormal area for subsequent inspection and processing.

[0111] The above steps can all be executed by a server or other device.

[0112] As can be seen from the above, the abnormality detection method of the new energy vehicle PCB board provided by the embodiment of the present application uses a multi-modal fusion network to extract and fuse the PCB board images of multiple different modes, which can integrate the multi-modal image data of the new energy vehicle PCB board, and perform abnormal area detection of the PCB board based on the features of the multi-modal image, which can improve the accuracy and efficiency of the detection. The embodiment of the present application also takes into account the specific requirements of the new energy vehicle PCB for wiring density and component integration, extracts related features from the multi-modal image features, and performs abnormality detection of the PCB board based on the features. It can be seen that the scheme can adapt to the special wiring density and component integration of the new energy vehicle PCB, and can further identify and locate the small abnormal defects and potential abnormalities on the new energy vehicle PCB under the requirements of wiring density and component integration, further improve the accuracy of abnormality detection of the new energy vehicle PCB board, and thus improve the reliability and safety of the new energy vehicle.

[0113] The following is a detailed description of the abnormality detection device for the new energy vehicle PCB board in this application. Please refer to Figure 5 . Figure 5 This is a schematic diagram of an embodiment of an abnormality detection device 50 in an embodiment of the present application. The abnormality detection device 50 includes:

[0114] The acquisition module 501 is used to acquire a PCB image and at least one supplementary image of the PCB board of the new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image;

[0115] The feature extraction module 502 is used to extract features from the PCB image and the supplementary image using a multimodal fusion network to obtain multimodal image features of the new energy vehicle PCB board;

[0116] A feature reconstruction module 503 is used to perform feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features;

[0117] The attention feature module 504 is used to extract attention features from the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board;

[0118] The distance acquisition module 505 is used to perform feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain the new energy vehicle PCB board feature distance image, where the PCB board sample image is the image feature of the sample PCB board based on the design parameters of the new energy vehicle PCB board, and the design parameters are parameters focusing on the standard wiring density and component integration of the new energy vehicle PCB board;

[0119] The detection module 506 is used to detect the abnormal area of ​​the new energy vehicle PCB board based on the characteristic distance image of the new energy vehicle PCB board and output the position information of the abnormal area.

[0120] In one embodiment, the multimodal fusion network includes: a plurality of input branch networks, a feature fusion layer, and a deep learning core layer, wherein each input branch corresponds to image feature extraction of one modality; a feature extraction module 502, which is used to:

[0121] The image corresponding to each modality is input into the branch network for feature extraction to obtain image features of different modalities;

[0122] A feature fusion layer is used to map the image features corresponding to different modalities into a common feature space, and the mapped features are fused in the common feature space to obtain fused features;

[0123] The deep learning core layer is used to perform deep feature extraction on the fused features and the multimodal image features of the PCB board of new energy vehicles.

[0124] In one embodiment, the feature reconstruction module 503 is used to:

[0125] Perform dimensionality reduction on multimodal image features;

[0126] According to the wiring density and component integration of new energy vehicle PCB boards, a deep convolutional autoencoder network is used to remove redundant information and retain key features of multimodal image features after dimensionality reduction;

[0127] The compressed features are upsampled and reconstructed through the decoder part of the deep convolutional autoencoder network to obtain the reconstructed multimodal image features.

[0128] The feature reconstruction module 503 is used to upsample and reconstruct the compressed features through the decoder part of the deep convolutional autoencoder network, wherein the decoder part includes multiple decoding layers, each decoding layer gradually restores the spatial resolution of the features through deconvolution operations or transposed convolution operations, so as to enhance the representation of the wiring density and component integration characteristics of the new energy vehicle PCB board, and obtain the reconstructed multimodal image features.

[0129] The characteristic distance image of the new energy vehicle PCB board includes pixel features; the detection module 506 is used to:

[0130] Extract the pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image;

[0131] The pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image are processed to obtain abnormal probability information of each pixel unit in the PCB image;

[0132] Based on the abnormal probability information of each pixel unit in the PCB image, the abnormal area of ​​the new energy vehicle PCB board is detected and the position information of the abnormal area is output.

[0133] In one embodiment, the detection module 506 is used to: set the pixel feature corresponding to each pixel unit in the PCB board image feature as a key, set the pixel feature corresponding to each pixel unit in the new energy vehicle PCB board feature distance image as a value, and perform feature processing of the mutual attention mechanism on the pixels corresponding to the pixel units in the new energy vehicle PCB board feature distance image to obtain the final pixel feature;

[0134] The abnormal probability information of each pixel unit in the PCB board image is calculated based on the final pixel features.

[0135] In one embodiment, the detection module 506 is used to:

[0136] Determine a candidate abnormal pixel unit in the PCB image based on abnormal probability information of each pixel unit in the PCB image;

[0137] Obtaining a distance threshold representing an abnormal pixel unit;

[0138] Compare the characteristic distance between each candidate abnormal pixel unit in the PCB board image and the corresponding pixel unit in the PCB board sample image with the distance threshold, and determine the abnormal result of each candidate abnormal pixel unit in the PCB board image according to the comparison result;

[0139] An abnormal area is determined in the PCB board image according to the abnormal result of each candidate abnormal pixel unit, and position information of the abnormal area is output.

[0140] The present application provides an anomaly detection device that can integrate multimodal image data of a new energy vehicle PCB board, thereby improving the accuracy and efficiency of detection; and the solution can adapt to the special wiring density and component integration of the new energy vehicle PCB, and can further identify and locate minor defects and potential anomalies on the new energy vehicle PCB, thereby improving the reliability and safety of the new energy vehicle.

[0141] Figure 6It is a schematic diagram of the structure of a server (computer device) provided in an embodiment of the present application. The server 300 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (CPU) 321 (for example, one or more processors) and memory 332, one or more storage media for storing application programs or data (for example, one or more mass storage devices). Among them, the memory 332 and the storage medium can be short-term storage or permanent storage. The program stored in the storage medium may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 321 can be configured to communicate with the storage medium and execute a series of instruction operations in the storage medium on the server 300.

[0142] The steps performed by the server in the above embodiment can be based on the Figure 6 The server structure shown. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices and units can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0143] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0146] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.

[0147] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application 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 replace some of the technical features therein by equivalents. 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 application.

Claims

1. A method for detecting abnormality of a PCB board of a new energy vehicle, characterized in that: include: Collecting a PCB image and at least one supplementary image of a PCB board of a new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image; Using a multimodal fusion network to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board; Performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features; Performing attention feature extraction on the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board; Performing feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, wherein the PCB board sample image is an image feature of a sample PCB board based on design parameters of the new energy vehicle PCB board, and the design parameters are parameters focusing on standard wiring density and component integration of the new energy vehicle PCB board; Detecting an abnormal area of ​​the new energy vehicle PCB board based on the characteristic distance image of the new energy vehicle PCB board and outputting position information of the abnormal area; The new energy vehicle PCB board characteristic distance image includes pixel features and characteristic distances between each pixel unit in the PCB board image and a corresponding pixel unit in the PCB board sample image; the detecting an abnormal area of ​​the new energy vehicle PCB board based on the new energy vehicle PCB board characteristic distance image and outputting position information of the abnormal area includes: Extracting the pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image; The pixel feature corresponding to each pixel unit in the PCB board image feature is set as a key, the pixel feature corresponding to each pixel unit in the new energy vehicle PCB board feature distance image is set as a value, and the pixel corresponding to the pixel unit in the new energy vehicle PCB board feature distance image is subjected to feature processing of a mutual attention mechanism to obtain a final pixel feature; and the abnormal probability information of each pixel unit in the PCB board image is calculated according to the final pixel feature; Based on the abnormal probability information of each pixel unit in the PCB image, candidate abnormal pixel units in the PCB image are determined; a distance threshold characterizing the abnormality of the pixel unit is obtained; a characteristic distance between each candidate abnormal pixel unit in the PCB board image and a corresponding pixel unit in the PCB board sample image is compared with the distance threshold, and an abnormal result of each candidate abnormal pixel unit in the PCB board image is determined according to the comparison result; an abnormal area is determined in the PCB board image according to the abnormal result of each candidate abnormal pixel unit, and position information of the abnormal area is output.

2. The anomaly detection method according to claim 1, characterized in that: The multimodal fusion network includes: a plurality of input branch networks, a feature fusion layer and a deep learning core layer, wherein each of the input branches corresponds to image feature extraction of one modality; A multimodal fusion network is used to extract features from the PCB image and the supplementary image to obtain multimodal image features of the new energy vehicle PCB board, including: The image corresponding to each modality is input into the branch network for feature extraction to obtain image features of different modalities; A feature fusion layer is used to map the image features corresponding to different modalities into a common feature space, and the mapped features are fused in the common feature space to obtain fused features; A deep learning core layer is used to perform deep feature extraction on the fused features, and multimodal image features of the new energy vehicle PCB board.

3. The anomaly detection method according to claim 1, characterized in that: The performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features includes: Performing dimensionality reduction processing on the multimodal image features; According to the wiring density and component integration of the new energy vehicle PCB board, a deep convolutional autoencoder network is used to remove redundant information and retain key features of the multimodal image features after dimensionality reduction; The compressed features are upsampled and reconstructed through the decoder part of the deep convolutional autoencoder network to obtain the reconstructed multimodal image features.

4. The anomaly detection method according to claim 3, characterized in that: The compressed features are upsampled and reconstructed through the decoder part of the deep convolutional autoencoder network to obtain the reconstructed multimodal image features, including: The compressed features are upsampled and reconstructed through the decoder part of the deep convolutional autoencoder network, where the decoder part includes multiple decoding layers, each of which gradually restores the spatial resolution of the features through deconvolution operations or transposed convolution operations to enhance the representation of the wiring density and component integration characteristics of the new energy vehicle PCB board, and obtain the reconstructed multimodal image features.

5. An abnormality detection device for a new energy vehicle PCB board, characterized in that: include: An acquisition module, used to acquire a PCB image and at least one supplementary image of a PCB board of a new energy vehicle, wherein the PCB image is a visible light image, and the supplementary image is an image of a different modality from the PCB image, including at least an infrared image and / or an X-ray image; A feature extraction module, used to extract features from the PCB image and the supplementary image using a multimodal fusion network to obtain multimodal image features of the new energy vehicle PCB board; A feature reconstruction module, used for performing feature reconstruction processing on the multimodal image features to obtain reconstructed multimodal image features; An attention feature module is used to extract attention features from the reconstructed multimodal image features to obtain PCB board image features of the wiring density and component integration of the new energy vehicle PCB board; A distance acquisition module is used to perform feature distance calculation processing on the PCB board image features and the PCB board sample image features to obtain a new energy vehicle PCB board feature distance image, wherein the PCB board sample image is an image feature of a sample PCB board based on the design parameters of the new energy vehicle PCB board, and the design parameters are parameters focusing on the standard wiring density and component integration of the new energy vehicle PCB board; A detection module, used to detect an abnormal area of ​​the new energy vehicle PCB board based on the characteristic distance image of the new energy vehicle PCB board and output position information of the abnormal area; The new energy vehicle PCB board feature distance image includes pixel features and a feature distance between each pixel unit in the PCB board image and a corresponding pixel unit in the PCB board sample image; The detection module is specifically used to extract the pixel features corresponding to each pixel unit in the PCB board image features and the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image; set the pixel features corresponding to each pixel unit in the PCB board image features as a key, set the pixel features corresponding to each pixel unit in the new energy vehicle PCB board feature distance image as a value, perform feature processing of the mutual attention mechanism on the pixels corresponding to the pixel units in the new energy vehicle PCB board feature distance image to obtain the final pixel features; calculate the abnormal probability information of each pixel unit in the PCB board image according to the final pixel features; determine the candidate abnormal pixel units in the PCB image based on the abnormal probability information of each pixel unit in the PCB image; obtain the distance threshold representing the abnormality of the pixel unit; compare the feature distance between each candidate abnormal pixel unit in the PCB board image and the corresponding pixel unit in the PCB board sample image with the distance threshold, and determine the abnormal result of each candidate abnormal pixel unit in the PCB board image according to the comparison result; determine the abnormal area in the PCB board image according to the abnormal result of each candidate abnormal pixel unit, and output the position information of the abnormal area.

6. A computer device, characterized in that: include: A memory and a processor; wherein the memory is used to store programs; The processor is used to execute the program in the memory, including executing the abnormality detection method for the new energy vehicle PCB board as described in any one of claims 1 to 4.

7. A computer-readable storage medium, comprising instructions, which, when executed on a computer, enables the computer to execute the method for detecting abnormalities of a PCB board of a new energy vehicle as claimed in any one of claims 1 to 4.

Images