A visual and electrical combined terminal wiring detection method

Abstract

The application relates to a terminal wiring detection method combining vision and electricity, which comprises the following steps: acquiring a to-be-detected image, pre-processing the to-be-detected image to obtain a pre-processed image; taking the pre-processed image as the input of a character detection and recognition deep learning model to acquire one-to-one corresponding position information and character information; extracting mutually matched terminal fixed labels and wire end labels from the character information; taking the pre-processed image as the input of an instance segmentation deep learning model to obtain an instance segmentation result; matching terminal feature marks based on the instance segmentation result and the terminal feature labels, wherein the terminal feature marks comprise wire end marks and anonymous marks; measuring the resistance value between two wire end marks with a pairing relationship among all wire end marks and measuring the resistance value between two anonymous marks among all anonymous marks; and acquiring a final detection result. Compared with the prior art, the application has the advantages of comprehensive detection state and high detection efficiency.

Description

Technical Field

[0001] This invention relates to a method for detecting the status of wire connections, and more particularly to a method for detecting terminal wiring that combines visual and electrical methods. Background Technology

[0002] Terminal blocks are widely used in various electrical control systems, serving functions such as power supply and signal transmission between different parts of the system. This wiring is mostly done manually by electrical engineers, which is prone to problems such as incorrect wiring, missing connections, and loose connections. Once these problems occur, they may lead to malfunctions in parts of the control system. Therefore, it is necessary to inspect the condition of the terminal block wiring when necessary. However, manual inspection of terminal block wiring is labor-intensive, inefficient, and prone to false positives. Therefore, an automated terminal block wiring inspection method is essential.

[0003] Currently, there are two different methods for inspecting terminal wiring: one based on electrical characteristics and the other based on image vision. Terminal wiring has four states: correct wiring and continuity (state 1), correct wiring but non-conductive (state 2), incorrect wiring but continuous (state 3), and incorrect wiring and non-conductive (state 4). Incorrect wiring means that points A and B should be connected, but are incorrectly connected to AC or other points; non-conductive means that there is an open circuit between any two points of the connecting wire. Neither image vision-based nor electrical characteristic-based inspection methods can completely detect all four states. Specifically, each single inspection method has the following characteristics and limitations:

[0004] Image-based detection methods rely on the requirement of visible labels at both ends of the wires connecting the two endpoints. Character detection and recognition algorithms identify these labels and match them to determine the correctness of the connection. For example, patent application CN112633343A discloses a method for verifying the wiring of a power equipment terminal block, comprising: receiving a physical diagram of the terminal block wiring of the power equipment; preprocessing the diagram; inputting the preprocessed diagram into a pre-trained target detection neural network model, which outputs terminal block position information; cropping terminal block diagrams based on the position information; inputting the terminal block diagrams into a text detection and recognition model, which outputs text block position information and a string; aligning the external wiring label with the terminal port number using an alignment strategy based on the text block position information and the string, forming structured information; and comparing the structured information with pre-stored wiring design schematic information corresponding to the terminal block to generate a verification result. However, image-based detection methods have the following drawbacks:

[0005] 1. It can only determine whether the wires are connected correctly, but cannot detect whether the wires are conductive;

[0006] 2. When the wire markings are obstructed and cannot be seen, the test may fail.

[0007] 3. It is susceptible to the influence of factors such as lighting environment, which may cause the label to become unrecognizable.

[0008] The method based on electrical characteristic detection is characterized by detecting the resistance between any two points by closing an external switch, and then determining whether there is continuity between the two points based on the resistance value. However, this method has the following drawbacks:

[0009] 1. It can only determine whether the wire is conductive. When the wire is not conductive, it cannot detect incorrect wire connections.

[0010] 2. When there are many connection points, the testing time will increase exponentially, becoming excessively long.

[0011] In addition, both of the above methods are currently limited to detecting the correctness of wire connections (whether the wire connections are correct and whether they are conductive), but cannot detect the standardization of wire connections. Summary of the Invention

[0012] The purpose of this invention is to overcome the shortcomings of the prior art and provide a visual and electrical combined terminal wiring inspection method that provides comprehensive and efficient detection of the status.

[0013] The objective of this invention can be achieved through the following technical solutions:

[0014] A method for detecting terminal wiring using both visual and electrical methods, comprising:

[0015] The image preprocessing step involves acquiring the image to be detected, performing preprocessing on the image to be detected, and obtaining a preprocessed image.

[0016] The character detection and recognition step uses the preprocessed image as input to the character detection and recognition deep learning model to obtain one-to-one corresponding position information and character information;

[0017] The character matching step involves extracting the terminal fixing labels and line end labels that match each other from the character information.

[0018] The instance segmentation step uses the preprocessed image as input to the instance segmentation deep learning model to obtain instance segmentation results, which include various wiring norm categories and their confidence levels.

[0019] The location marking step involves matching and obtaining terminal feature marks based on the instance segmentation results and the terminal feature labels. The terminal feature marks include line end marks and anonymous marks.

[0020] The resistance measurement procedure involves measuring the resistance between any two paired line end marks and measuring the resistance between any two anonymous marks among all anonymous marks.

[0021] The detection step involves obtaining the final detection result based on the resistance value and the instance segmentation result.

[0022] Furthermore, the preprocessing specifically includes:

[0023] First, the RGB channels of the image to be detected are normalized. Then, the w and h channels of the image are expanded upward to a set length. Finally, the image is converted into tensor form to obtain the preprocessed image.

[0024] Furthermore, the deep learning model for character detection and recognition includes:

[0025] The backbone network used for feature map extraction from preprocessed images;

[0026] A character detection network is obtained based on the extracted feature maps to obtain the position information of character detection boxes;

[0027] Based on the position information of the character detection box, the character feature information of the corresponding position in the feature map is cropped, and the character recognition network of the character at the corresponding position is obtained based on the character feature information.

[0028] Furthermore, the character matching step specifically includes:

[0029] Match information that conforms to the terminal labeling rules from the character information, extract it from the character information as the fixed terminal label, and retrieve its position information;

[0030] Find information that matches the line end labeling rules from the character information, use it as the line end label, and extract its position information;

[0031] Traverse all line end labels, find the terminal fixed label with the shortest distance based on the position information, and save it as matching information.

[0032] Furthermore, if the distance between the fixed terminal labels that match the line end labels is greater than the shortest distance threshold, then the current fixed terminal label interval is abandoned.

[0033] Furthermore, the instance segmentation deep learning model includes:

[0034] The backbone network is used to extract feature maps from preprocessed images, wherein the feature maps are multiple feature maps of different scales;

[0035] An FPN fusion network is used to fuse information at different levels between multiple feature maps of different scales, thereby obtaining multiple feature maps of different sizes.

[0036] An instance class prediction network that generates instance class prediction outputs at multiple different stages based on the multiple feature maps of different sizes;

[0037] A mask prediction network is generated based on the multiple feature maps of different sizes to produce instance mask outputs at multiple different stages.

[0038] The output network combines the instance category prediction output and instance mask output to produce the instance segmentation result.

[0039] Furthermore, the matching to obtain the terminal feature marker specifically involves:

[0040] Detect the position of the line ends in the output of all instances of the mask, find the terminal fixed label closest to the line end, and match and count;

[0041] Iterate through the current instance mask and output the fixed terminal labels that are matched. If the number N of the paired line labels is less than the number M of the currently matched mask line labels, add MN anonymous tags to the current fixed terminal label.

[0042] Furthermore, when outputting instance segmentation results, if the confidence level is lower than a set threshold, the corresponding instance segmentation results are deleted.

[0043] Furthermore, the resistance is measured using a relay array, specifically by closing the relay switch corresponding to the mark and measuring the resistance between the two marks.

[0044] Furthermore, in the resistance measurement step, all line end marks are traversed, and if no matching mark exists, the mark is added to the anonymous mark.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] First, this invention is the first to propose a terminal detection method that combines image detection and electrical detection, which makes up for the shortcomings of a single detection method that can only detect a limited number of states. This invention can detect the state of all terminal wiring, including correct wiring and wiring continuity, correct wiring but wiring non-conductive, incorrect wiring but wiring continuity, incorrect wiring and wiring non-conductive, etc.

[0047] Second, the present invention uses an instance segmentation algorithm, which can detect not only the correctness of the terminal wiring, but also the standardization issues of the terminal wiring, mainly including five types of standardization issues: exposed metal at the wire end, bent and tangled wiring, missing markings on the wiring, severely skewed wire end, and detached wire end, totaling 32 categories.

[0048] Third, this invention achieves partial terminal wiring pairing through character detection and recognition, greatly reducing the number of relay array traversal measurements and saving overall testing time. Furthermore, the fault tolerance and accuracy of this invention are higher than those of single detection methods. Attached Figure Description

[0049] Figure 1 This is a flowchart of the overall detection process of the present invention;

[0050] Figure 2 Here is a flowchart of the character detection and recognition process;

[0051] Figure 3 Flowchart for character matching;

[0052] Figure 4 Flowchart for image instance segmentation;

[0053] Figure 5 Flowchart for location tagging information processing;

[0054] Figure 6 Here is a flowchart for relay array testing;

[0055] Figure 7 This is a schematic diagram for measuring the resistance of a relay array. Detailed Implementation

[0056] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0057] This invention provides a method for detecting terminal wiring that combines visual and electrical methods, such as... Figure 1 As shown, it includes the following steps:

[0058] S1. Obtain the image to be detected, and preprocess the image to be detected to obtain a preprocessed image.

[0059] In a specific implementation, the preprocessing is as follows:

[0060] First, the RGB channels of the image to be detected are normalized. Then, the w and h values ​​of the image are expanded upwards to a set length. Finally, the image is converted into tensor form to obtain the preprocessed image. Specifically, the w and h values ​​of the image are expanded upwards to a multiple of the nearest integer 32.

[0061] S2. Using the preprocessed image as input to the character detection and recognition deep learning model, obtain one-to-one corresponding position information and character information.

[0062] In a specific implementation, the deep learning model for character detection and recognition includes:

[0063] The backbone network used for feature map extraction from preprocessed images;

[0064] A character detection network is obtained based on the extracted feature maps to obtain the position information of character detection boxes;

[0065] Based on the position information of the character detection box, the character feature information of the corresponding position in the feature map is cropped, and the character recognition network of the character at the corresponding position is obtained based on the character feature information.

[0066] like Figure 2 As shown, the specific process of obtaining location information and character information includes the following steps:

[0067] S201. Determine whether to load the character detection and recognition deep learning model. If yes, proceed to step S202. If no, load the character detection and recognition deep learning model through initialization.

[0068] S202. Input the image tensor information into the backbone network, perform forward computation, extract feature information, and obtain the feature map.

[0069] S203. Input the extracted feature map into the character detection network to obtain the position information of the character detection box;

[0070] S204. Using the position information of the character detection box, crop out the character feature information at the corresponding position in the feature map, and input all the character feature information into the character recognition network to obtain the character at the corresponding position;

[0071] S205. Save the location information and character information in a one-to-one correspondence.

[0072] The aforementioned deep learning model for character detection and recognition needs to be pre-trained, specifically as follows:

[0073] a) Prepare a dataset for character detection and recognition in terminal wiring scenarios, consisting of approximately 1000 images, which need to be labeled with character position information and character labels;

[0074] b) Build an end-to-end FOTS character detection and recognition network model in the PyTorch deep learning framework, use SGD as the model optimizer, with a batch size of 8, and iterate and train on the GPU for about 300,000 times.

[0075] c) Save the trained model.

[0076] S3. Extract the matching terminal fixed labels and line end labels from the character information.

[0077] like Figure 3 As shown, the specific matching process includes the following steps:

[0078] S301, Detecting fixed terminal labels: Matching information that conforms to the terminal labeling rules from the character information, finding it from the character information as a fixed label, and extracting its position information;

[0079] S302. Detect the wire end markings: Find information that matches the wire end marking rules from the character information, use it as the wire end marking, and extract its position information;

[0080] S303. Matching line end labels with fixed labels: Traverse the line end labels and find the fixed label with the shortest distance to it, and save it as the matching information.

[0081] In step S303, if the distance between the fixed terminal labels that match the line end labels is greater than the shortest distance threshold, then the current fixed terminal label is abandoned. In a specific implementation, the shortest distance threshold can be 5cm. If it exceeds 5cm, it is considered that there is no matching object, and the current label is abandoned. Here, 5cm is an actual value, and the pixel size needs to be scaled according to the actual ratio.

[0082] S4. Using the preprocessed image as input to the instance segmentation deep learning model, the instance segmentation results are obtained, including various wiring standardization categories and their confidence levels.

[0083] In a specific implementation, the instance segmentation deep learning model includes:

[0084] The backbone network is used to extract feature maps from the preprocessed image. The feature maps are multiple feature maps of different scales. In this embodiment, four different scales are set.

[0085] An FPN fusion network is used to fuse information at different levels between multiple feature maps of different scales, thereby obtaining multiple feature maps of different sizes.

[0086] An instance class prediction network that generates instance class prediction outputs at multiple different stages based on the multiple feature maps of different sizes;

[0087] A mask prediction network is generated based on the multiple feature maps of different sizes to produce instance mask outputs at multiple different stages.

[0088] The output network combines the instance category prediction output and instance mask output to produce the instance segmentation result.

[0089] like Figure 4 As shown, the instance segmentation process includes the following steps:

[0090] S401. Determine whether the instance segmentation deep learning model has been loaded. If yes, proceed to step S402. If no, load the instance segmentation deep learning model by initialization. When using the model again, simply call the model.

[0091] S402, Feature Extraction: Input the image tensor information into the backbone network of the deep learning model to extract feature information and obtain different feature maps output in 4 stages;

[0092] S403, FPN network fusion: The FPN network is used to fuse information at different levels between the inputs of the four stages, and finally outputs four fused feature maps of different sizes.

[0093] S404, Instance Class Prediction: Input four fused feature maps of different sizes into the instance class prediction network to obtain class prediction outputs at four different stages;

[0094] S405, Example Mask Prediction: Input four fused feature maps of different sizes into the Mask Prediction Network to obtain four different stage mask outputs;

[0095] S406. Joint Result Output: The instance segmentation result is output by combining the instance category prediction results and the instance mask prediction results. Finally, the result is saved.

[0096] In a preferred embodiment, when outputting instance segmentation results, if the confidence level is lower than a set threshold, the corresponding instance segmentation result is deleted. The set threshold can be 0.1.

[0097] In this embodiment, the instance segmentation deep learning model is a modified SOLOv2 instance segmentation model. The main difference is that the model's category output is changed to 32 wire canonicality problem categories. Furthermore, some convolutional operations are replaced by a transformer.

[0098] In this embodiment, the instance segmentation deep learning model needs to be pre-trained, specifically as follows:

[0099] a) Prepare a dataset of wire instance segmentation for terminal wiring scenarios, approximately 2000 images. Different wire instances need to be labeled one by one, and the problem type of the current wire needs to be marked.

[0100] b) Build an instance segmentation network model in the PyTorch deep learning framework, use SGD as the optimizer, with a batch size of 8, and iterate 300,000 times on the GPU.

[0101] c) Save the trained instance segmentation network model.

[0102] S5. Based on the instance segmentation result and the terminal feature number, a terminal feature marker is obtained by matching. The terminal feature marker includes a line end marker and an anonymous marker.

[0103] like Figure 5 As shown, the matching tagging process includes the following steps:

[0104] S501. Detect the position of the line end in the output of all instances of mask, find the terminal fixed label closest to the line end, and match and count;

[0105] S502. Traverse the current instance mask and output the matched terminal fixed labels. If the number of matched line terminal labels N is less than the number of currently matched mask line terminals M, add MN anonymous tags to the current terminal fixed label.

[0106] In this embodiment, the so-called anonymous label specifically refers to: the terminal at the fixed label is connected to a wire that is not labeled or whose label recognition fails. These wires do not have distinguishable markings, so they are marked with indistinguishable anonymous labels.

[0107] S6. Measure the resistance between any two paired line end marks, and measure the resistance between any two anonymous marks in all anonymous marks.

[0108] In a specific implementation, resistance is measured using a relay array. Specifically, the relay switch corresponding to the marked point is closed, and the resistance between the two marks is measured. Specifically, a relay is connected to the outside of the terminal block. When it is necessary to measure the resistance between points A and B, the relay contacts around points A and B are closed, and the resistance between them is measured. The closing and opening of the relay contacts can be controlled by a PLC.

[0109] like Figure 6 As shown, measuring resistance using a relay array specifically includes the following steps:

[0110] S601. Measure the resistance between matching marks: Traverse all line end marks. If a matching mark exists, take out the matching mark, close the relay switch corresponding to the mark, and measure the resistance between them. If no matching mark exists, add the mark to the anonymous mark until all line end marks have been traversed.

[0111] S602. Measure the resistance between anonymous tokens: Traverse all anonymous tokens and measure the resistance between each token and all other anonymous tokens until the traversal ends.

[0112] S7. Obtain the final test result based on the resistance value and the instance segmentation result, and obtain the test result information on the correctness and standardization of the wiring. Compare with the correct wiring table, and output the information on the currently correct wiring and the information on incorrect wiring. The information mainly includes: the fixed label of the wiring, the terminal label (if none, it will not be output), whether it is correctly connected, the measured resistance value, and the category of non-standard wiring.

[0113] This invention can detect the status of all terminal wiring through the above method. In addition to detecting the correctness of the terminal wiring, it can also detect the standardization issues of the terminal wiring, mainly including five types of standardization issues: exposed metal at the wire end, bent and tangled wire, missing markings on the wire, severely skewed wire end, and detached wire end, totaling 32 categories. The detection is comprehensive, reliable, and efficient.

[0114] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0115] In one embodiment, the correct wiring is terminal 1 connected to terminal N+1, and the actual wiring is terminal 1 connected to terminal N+1, such as... Figure 7As shown. An image of the wiring at the terminals is obtained. The character detection and recognition model identifies the fixed label 1 at terminal 1, the fixed label N+1 at terminal N+1, the wire end label 1 of the wiring at terminal 1, and the wire end label N+1 of the wiring at terminal N+1 (all labels appearing in the image are identified). Fixed label 1 at terminal 1 matches wire end label 1, and fixed label N+1 at terminal N+1 matches wire end label N+1. Since wire end label 1 and wire end label N+1 are on the same wire, a match exists, indicating a connection between terminal 1 and terminal N+1. The instance segmentation model predicts the wiring standardization error category, segmenting the wire instances in the image. Terminal 1 matches one masked wire end with no anonymous marker, and terminal N+1 matches one masked wire end with no anonymous marker. When measuring resistance, relay contact K_1 closes, relay contact K_N+1 closes, the resistance between terminals is measured, and the contacts open after the measurement is complete. Final output: Terminal 1 and terminal N+1 are correctly wired, the resistance value is within the specified range, the wiring is continuous, and there are no standard wiring errors (if any, output the error type).

[0116] In one embodiment, the correct wiring is: terminal 1 connected to terminal N+1, terminal 1 connected to terminal N+2. The actual wiring is: terminal 1 connected to terminal N+1 (wire wiring label 1_1 at terminal 1 is visible), terminal 1 connected to terminal 2N (wire wiring label 1_2 at terminal 1 is not visible due to obstruction), and there is a standard error at terminal 2N: "exposed metal wire end". By obtaining images of the wiring at the terminals, the character detection and recognition model identifies the fixed label 1 at terminal 1, the fixed label N+1 at terminal N+1, the fixed label N+2 at terminal N+2, the fixed label 2N at terminal 2N, the visible terminal label 1_1 at terminal 1, the obstructed terminal label 1_2 at terminal 1, the terminal label N+1 at terminal N+1, and the terminal label 2N at terminal 2N. At terminal 1, fixed label 1 matches wire end label 1_1; at terminal N+1, fixed label N+1 matches wire end label N+1; at terminal 2N, fixed label 2N matches wire end label 2N. Since wire end labels 1_1 and N+1 are on the same wire and match, a connection exists between terminals 1 and N+1. The instance segmentation model segments wire instances in the image and predicts the wiring standardization error category. At terminal 1, there are two masked wire ends, greater than 1 (one matching wire end label 1_1 is identified), therefore, fixed label 1 is incremented by an anonymous marker {anonymous marker at terminal 1}. At terminal N+1, there is one masked wire end and one wire end label, so no anonymous marker is added. At terminal 2N, there is one masked wire end and one wire end label, so no anonymous marker is added. Furthermore, the wire at terminal 2N is predicted to have a "bare wire end" standardization error. When measuring resistance, first iterate through the terminal labels {terminal label 1_1, terminal label N+1, terminal label 2N}. Terminal label 1_1 and terminal label N+1 have a match. Close relay contacts K_1 and K_N+1 to directly measure the resistance between terminal 1 and terminal N+1. Terminal label 2N has no matching label, so it is added to the anonymous label set {anonymous label at terminal 1, anonymous label at terminal 2N}. Iterate through this set of anonymous labels and measure the resistance between each pair of labels. Compare with the correct wiring list; the resistance between terminal 1 and terminal N+2 also needs to be measured. Final output: Terminal 1 and terminal N+1 are correctly wired, the resistance is within the specified range, the wiring is continuous, and there is no standard wiring error (if there is, output the error type); Terminal 1 and terminal 2 are not wired, the resistance is infinite, the wiring is not continuous, and there is no standard output; Terminal 1 and terminal 2N are wired, which is an incorrect wiring, the resistance is within the specified range, the wiring is continuous, and there is a standard error of "exposed metal wire end".

[0117] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for detecting terminal wiring using a combination of visual and electrical methods, characterized in that, include: The image preprocessing step involves acquiring the image to be detected, performing preprocessing on the image to be detected, and obtaining a preprocessed image. The character detection and recognition step uses the preprocessed image as input to the character detection and recognition deep learning model to obtain character information and its corresponding position information; The character matching step involves extracting the terminal fixing labels and line end labels that match each other from the character information. The instance segmentation step uses the preprocessed image as input to the instance segmentation deep learning model to obtain instance segmentation results, which include various wiring norm categories and their confidence levels. The location marking step involves matching and obtaining terminal feature marks based on the instance segmentation results and the terminal feature labels. The terminal feature marks include line end marks and anonymous marks. The resistance measurement procedure involves measuring the resistance between any two paired line end marks and measuring the resistance between any two anonymous marks among all anonymous marks. The detection step involves obtaining the final detection result based on the resistance value and the instance segmentation result.

2. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, The preprocessing specifically includes: First, the RGB channels of the image to be detected are normalized. Then, the width and height of the image are expanded upward to a set length. Finally, the image is converted into tensor form to obtain the preprocessed image.

3. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, The deep learning model for character detection and recognition includes: The backbone network used for feature map extraction from preprocessed images; A character detection network is obtained based on the extracted feature maps to obtain the position information of character detection boxes; Based on the position information of the character detection box, the character feature information of the corresponding position in the feature map is cropped, and the character recognition network of the character at the corresponding position is obtained based on the character feature information.

4. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, The character matching steps specifically include: Match information that conforms to the terminal labeling rules from the character information, extract it from the character information as the fixed terminal label, and retrieve its position information; Find information that matches the line end labeling rules from the character information, use it as the line end label, and extract its position information; Traverse all line end labels, find the terminal fixed label with the shortest distance based on the position information, and save it as matching information.

5. The terminal wiring detection method combining vision and electrical methods according to claim 4, characterized in that, If the distance between the fixed terminal labels that match the line end labels is greater than the shortest distance threshold, then the current fixed terminal label is discarded.

6. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, The instance segmentation deep learning model includes: The backbone network is used to extract feature maps from preprocessed images, wherein the feature maps are multiple feature maps of different scales; An FPN fusion network is used to fuse information at different levels between multiple feature maps of different scales, thereby obtaining multiple feature maps of different sizes. An instance class prediction network that generates instance class prediction outputs at multiple different stages based on the multiple feature maps of different sizes; A mask prediction network is generated based on the multiple feature maps of different sizes to produce instance mask outputs at multiple different stages. The output network combines the instance category prediction output and instance mask output to produce the instance segmentation result.

7. The terminal wiring detection method combining vision and electrical methods according to claim 6, characterized in that, The specific method for obtaining terminal feature markers through matching is as follows: Detect the position of the line ends in the output of all instances of the mask, find the terminal fixed label closest to the line end, and match and count; Iterate through the current instance mask and output the fixed terminal labels that are matched. If the number N of the paired line labels is less than the number M of the currently matched mask line labels, add MN anonymous tags to the current fixed terminal label.

8. The terminal wiring detection method combining vision and electrical methods according to claim 6, characterized in that, When outputting instance segmentation results, if the confidence level is lower than a set threshold, the corresponding instance segmentation results are deleted.

9. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, The resistance is measured using a relay array. Specifically, the relay switch corresponding to the mark is closed, and the resistance between the two marks is measured.

10. The terminal wiring detection method combining vision and electrical methods according to claim 1, characterized in that, In the resistance measurement step, all line end marks are traversed. If no matching mark exists, the mark is added to the anonymous mark.

Images