Component defect identification method and device of distribution network line, electronic equipment and medium

Abstract

The invention discloses a component defect identification method and device of a distribution network line, electronic equipment and a medium, and relates to the technical field of power system defect detection, the method is used for solving the problems that a model is weak in small target identification capability and low in identification precision, and the method comprises the following steps: obtaining a distribution network line image, inputting the distribution network line image into a target model; performing slicing processing on the distribution network line image based on the target model, and generating a prediction frame according to a plurality of slice images and the distribution network line image; the prediction frame comprises a plurality of defect prediction frames; and adopting a preset non-maximum suppression algorithm based on the area of the defect prediction frames and priori knowledge, and filtering in the plurality of defect prediction frames to obtain a plurality of defect target frames. By adopting the method, the small target identification capability of the model can be improved, the false detection rate of the model in a distribution network scene can be reduced, and the small target defect identification precision of the model can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system defect detection, and in particular to a method, device, electronic equipment and medium for identifying component defects of a distribution network line. Background Art

[0002] With the rapid advancement of deep learning and computer vision technology, drone inspection is a common application scenario in the power distribution network industry. In related technologies, the components in the images taken by drone inspections are relatively small, which makes it difficult to detect defective targets of components solely by models. In addition, there are factors in the images that interfere with model recognition, resulting in low recognition accuracy of defective targets of components. Summary of the invention

[0003] Based on this, it is necessary to provide a method, device, electronic device and medium for component defect identification of distribution network lines that can improve the model's recognition accuracy for defective targets in response to the above technical problems.

[0004] A method for identifying component defects of a distribution network line, comprising: Acquire a distribution network line image, and input the distribution network line image into a target model; Slicing the distribution network line image based on the target model, and generating a prediction frame according to a plurality of slice images and the distribution network line image; the prediction frame includes a plurality of defect prediction frames; A preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge are used to filter out multiple defect target frames from the multiple defect prediction frames.

[0005] In the above scheme, the preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame filters multiple defect target frames in multiple defect prediction frames, including: According to the area sizes of the multiple defect prediction boxes, the multiple defect prediction boxes are sorted in descending order; According to the sorted order, the area index between any two of the defect prediction frames is calculated in sequence; the area index represents the ratio between the overlapping area of ​​the two defect prediction frames and the minimum area of ​​the two defect prediction frames; If the area index is greater than a set filtering threshold, the defect prediction frame with a smaller area among the two defect prediction frames corresponding to the area index is filtered out.

[0006] In the above solution, the prediction frame also includes multiple category prediction frames, and the prior knowledge is used to filter multiple defect target frames in the multiple defect prediction frames, including: The plurality of defect prediction frames are filtered according to the category prediction frame; the category prediction frame is used to characterize the region of invalid defect distribution determined based on prior knowledge.

[0007] In the above solution, the category prediction box includes a non-main tower category prediction box and a discarded component category prediction box.

[0008] In the above solution, filtering the plurality of defect prediction frames according to the category prediction frame includes: If the category prediction frame is a non-main tower category prediction frame, filtering out the defect prediction frame located in the non-main tower category prediction frame; If the category prediction box is a discarded component category prediction box, it is calculated whether the discarded component category prediction box intersects with the defect prediction box, and the defect prediction box intersecting with the discarded component category prediction box is filtered out.

[0009] In the above solution, the slicing process of the distribution network line image based on the target model and generating a prediction frame according to a plurality of slice images and the distribution network line image includes: Setting cutting parameters, and slicing the distribution network line image based on the cutting parameters to obtain a plurality of sliced ​​images; the cutting parameters include: a width direction cutting ratio, a height direction cutting ratio, and a height and width overlapping area ratio; A plurality of the slice images and the distribution network line images are spliced ​​into a set of data samples.

[0010] In the above solution, generating a prediction frame according to the multiple slice images and the distribution network line image includes: According to the position correspondence between the plurality of slice images and the distribution network line image, the prediction boxes in the plurality of slice images are restored to the coordinates of the distribution network line image.

[0011] A device for identifying component defects of a distribution network line, comprising: An acquisition module, used for acquiring a distribution network line image and inputting the distribution network line image into a target model; An inference module, configured to slice the distribution network line image based on the target model, and generate a prediction frame according to a plurality of slice images and the distribution network line image; the prediction frame includes a plurality of defect prediction frames; The filtering module is used to adopt a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge to filter multiple defect target frames from the multiple defect prediction frames.

[0012] An electronic device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method for identifying component defects of a distribution network line when executing the computer program.

[0013] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for identifying component defects in a distribution network line.

[0014] The above-mentioned component defect recognition method, device, electronic device and medium of the distribution network line slices the distribution network line image based on the target model, generates multiple defect prediction frames based on multiple slice images and the distribution network line image, thereby improving the recognition accuracy of the model for component defects by introducing a slice reasoning scheme. Then, a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge are used to filter multiple defect target frames in multiple defect prediction frames, thereby effectively reducing the probability of model misdetection, and filtering out the prediction frames generated by the model under interference, further improving the recognition accuracy of the model for component defects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A schematic diagram of a flow chart of a method for identifying component defects of a distribution network line in one embodiment; Figure 2 The figure is a schematic diagram of a process of processing a distribution network line image by a target model in one embodiment; Figure 3 is a schematic diagram of overlapping defect prediction boxes in one embodiment; Figure 4 A schematic diagram of a flow chart of a preset non-maximum suppression algorithm in one embodiment; Figure 5 The present invention is a structural block diagram of a device for identifying component defects in a distribution network line in an embodiment. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] The implementation details of the technical solution of the embodiment of the present invention are described in detail below.

[0018] In one embodiment, Figure 1 As shown, a method for identifying component defects of a distribution network line is provided, and the method may include the following steps: Step S101, obtaining a distribution network line image, and inputting the distribution network line image into a target model.

[0019] Distribution network lines refer to the line system used to transmit electricity. Distribution network line images refer to images used to display the lines, poles, equipment and related components in the power distribution network system, which can be obtained through aerial photography by distribution network drones.

[0020] Distribution network drone inspection refers to the use of drones to inspect and monitor the power distribution network system. During the inspection process, it is ensured that the distribution network drone can fly over the power distribution network line and take pictures of the distribution network line. Among them, the distribution network line image includes the components on the tower, so that it is possible to identify whether the components on the tower have defects from the distribution network line image.

[0021] In practical applications, the acquired distribution network line images are preprocessed, including noise removal, brightness and contrast adjustment, etc., to improve image quality, ensure image clarity and accuracy, and facilitate subsequent model analysis and recognition.

[0022] Then, the distribution network line image is input into the target model for identification and analysis. The target model is designed and trained for analyzing the distribution network line image. The target model can identify the component defects in the distribution network line image after a series of processing and analysis.

[0023] Step S102, slicing the distribution network line image based on the target model, and generating a prediction box according to a plurality of slice images and the distribution network line image.

[0024] The input distribution network line image is obtained through the inspection of the distribution network drone. The components on the distribution network line image are small, and the defects on the components are even smaller. Based on this, in order to enable the target model to have a better recognition ability for small targets, the target model introduces a slice reasoning mechanism, so that it can better identify component defects in the distribution network line image. Among them, the slice reasoning mechanism is specifically: based on the target model, the distribution network line image is sliced ​​to obtain multiple slice images, where there are overlapping parts between adjacent slice images, and multiple slice images are spliced ​​together to obtain a complete distribution network line image.

[0025] The sliced ​​image obtained after the slicing process contains information about a specific area, and the distribution network line image contains information about the entire area. Therefore, the target model performs reasoning based on multiple sliced ​​images and the input distribution network line image to generate multiple prediction boxes. Among them, the prediction box includes multiple defect prediction boxes, which are used to indicate components that may have defects in the distribution network line image.

[0026] In this embodiment, by introducing a slice inference mechanism, the distribution network line image is sliced ​​into multiple slice images, which can make the detection object in each slice image larger than the entire image. This can increase the size of the detection object, so that the detection object is easier to be detected and recognized by the target model. The slice image contains local information of the detection object, which helps the model to better learn the details and local features of the detection object and improve the model's recognition ability for small targets.

[0027] In one embodiment, Figure 2 As shown, Figure 2 A schematic diagram of the target model's processing flow for distribution network line images is shown.

[0028] Step S201, setting cutting parameters, and slicing the distribution network line image based on the cutting parameters to obtain a plurality of slice images.

[0029] In the process of slicing, you first need to set the cutting parameters, which determine how to divide the distribution network line image into multiple slice images. Specifically, it includes setting the width direction cutting ratio to , Height direction cutting ratio , height and width overlap area ratio Among them, the width-direction cutting ratio refers to the ratio of the width of each slice image to the width of the original image when the image is cut in the horizontal direction; the height-direction cutting ratio refers to the ratio of the height of each slice image to the height of the original image when the image is cut in the vertical direction; the height and width overlapping area ratio refers to the ratio of the width and height of the overlapping area of ​​two adjacent slice images in the width and height directions of the slice image.

[0030] According to the set cutting parameters, the distribution network line image is sliced. In one implementation, the size of the slice window is set according to the width cutting ratio and the height cutting ratio, thereby limiting the size of the slice image. Then the slice window is moved on the distribution network line image and small slice images are intercepted one by one. In this process, there will be overlapping areas between each slice image in the height and width directions. The size of this overlapping area is determined by the parameter Control, for example, if you set the parameter , then there will be 20% pixel overlap between two adjacent slice images in the height and width directions, which means that in the slice image after cutting, some pixels will belong to two adjacent slice images at the same time, which helps to obtain more context information and ensure the continuity of image slices.

[0031] Assume that the size of the input distribution network line image A is After slicing, the distribution network line image A can be cut into That is, it is divided into 3 slice images in the horizontal direction and 2 slice images in the vertical direction, and a total of 6 slice images are obtained. These 6 slice images are , where, according to the above-set cutting parameters and slicing processing method, the coordinates of the slice image on the distribution network line image are expressed as:

[0032]

[0033]

[0034]

[0035]

[0036]

[0037] Thus, six segmented images on the distribution network line image are obtained.

[0038] Step S202: splice multiple slice images and distribution network line images into a group of data samples.

[0039] The distribution network line image and multiple slice images obtained by slicing are spliced ​​into a set of data samples (batch) with a length of 7, expressed as Then, the target model will perform model inference on this set of data samples and generate the corresponding prediction box.

[0040] In one embodiment, the target model performs model inference on the distribution network line image and multiple slice images, that is, the target model will give the prediction boxes corresponding to the distribution network line image and each slice image. Since the sizes of the distribution network line image and the slice image are different, the prediction boxes on the slice image and the prediction boxes on the distribution network line image are not in a unified scale range. Based on this, it is necessary to normalize the prediction boxes on the slice image to the coordinate system of the distribution network line image so that each prediction box is in a unified scale range.

[0041] Here, according to the position correspondence between the distribution network line image and the slice image, the coordinates of the prediction box on the slice image are mapped to the coordinates corresponding to the distribution network line image, so as to obtain the position of the prediction box on the slice image on the distribution network line image, so that the target model finally outputs the prediction box in the distribution network line image coordinate system.

[0042] Step S103 , using a preset non-maximum suppression algorithm based on the area of ​​the defect prediction box and prior knowledge, filtering multiple defect prediction boxes to obtain multiple defect target boxes.

[0043] Non-maximum suppression can filter out some repeated defect prediction boxes. Figure 3 As shown, Figure 3 A schematic diagram of overlapping defect prediction boxes is shown. Figure 3 For the overlapping defect prediction boxes shown in the figure, if the conventional non-maximum suppression processing flow is used to filter the repeated prediction boxes, the filtering threshold needs to be set to a smaller value. However, in the distribution network scenario, many targets have large overlapping areas. If the filtering threshold is set to a smaller value, the target will be missed.

[0044] Based on this, this embodiment proposes a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame. In the process of filtering overlapping defect prediction frames, the preset non-maximum suppression algorithm takes into account the area size of the defect prediction frame, and can distinguish the overlap between larger defect target frames and smaller defect target frames. Therefore, the preset non-maximum suppression algorithm supports setting a higher filtering threshold, so that defect prediction frames with smaller areas can be filtered out by the set higher filtering threshold, retaining defect prediction frames with larger areas or relatively independent ones, and at the same time will not cause missed detection of targets in other scenarios.

[0045] In practical applications, distribution network line images often contain information about multiple towers, among which the main tower is the main supporting structure in the power distribution network system, connecting high-voltage transmission lines and distribution lines, and assuming the bearing and transmission functions of transmission lines and distribution lines. Based on this, what actually needs to be identified is the component defects in the main tower information. At the same time, considering that there are many discarded components on the towers, the distribution network defects on the discarded components do not need to be identified and counted. The multiple defect prediction boxes generated by the target model inference include invalid defect prediction boxes. These invalid defect prediction boxes are obtained by mistake when the target model is interfered by non-main tower information and discarded components. Therefore, it is necessary to exclude the defect prediction boxes about other tower information and discarded components from the multiple defect prediction boxes.

[0046] Based on this, the prior knowledge of the distribution network scenario is introduced to filter multiple defect prediction frames. The prior knowledge of the distribution network scenario can indicate the invalid defect prediction frames that need to be filtered out, thereby filtering out the invalid defect prediction frames in multiple defect prediction frames, which can reduce the false detection rate of the target model and reduce the interference of other tower information and waste materials in the distribution network line image on model recognition. In addition, by introducing prior knowledge to filter out multiple defect prediction frames, the workload of staff for reviewing the defect target frame can be reduced after the final defect target frame is generated.

[0047] After filtering multiple defect prediction frames using the preset non-maximum suppression algorithm and prior knowledge, the retained defect prediction frames are multiple defect target frames. These defect target frames are the defect recognition results finally output by the target model, which are used to mark the defects of effective components on the main tower. They do not contain overlapping target frames and falsely detected target frames, thereby improving the target model's ability to recognize component defects.

[0048] In one embodiment, Figure 4 As shown, Figure 4 A schematic flow chart of a preset non-maximum suppression algorithm is shown.

[0049] Step S401 , sorting the multiple defect prediction boxes in descending order according to the area sizes of the multiple defect prediction boxes.

[0050] Here, large-area defect prediction boxes usually cover more areas, may contain more target information, and are more likely to be the main target, while small-area defect prediction boxes are usually noise or multiple targets. Based on this, the defect prediction boxes are sorted in descending order according to their area size. The sorting results can be used to consider the area size of the defect prediction boxes, which helps to better control the overlap between defect prediction boxes.

[0051] Step S402, calculating the area index between any two defect prediction frames in sequence according to the sorted order.

[0052] In the sorted defect prediction box list, each defect prediction box is traversed, and the area index between the defect prediction box in the current order and other defect prediction boxes is calculated in turn. This area index is used to measure the degree of overlap between two defect prediction boxes, and it changes with the area size and overlap degree of the two defect prediction boxes. Therefore, this area index further considers the impact of the area size of the defect prediction box on the overlap degree.

[0053] The area index is obtained by calculating the ratio between the overlapping area of ​​two defect prediction boxes and the minimum area of ​​the two defect prediction boxes. Defect prediction box The area index between The calculation process is:

[0054] in, Defect prediction box Defect prediction box The overlapping area between Defect prediction box Defect prediction box The minimum value of the area.

[0055] Step S403: if the area index is greater than the set filtering threshold, the defect prediction frame with a smaller area among the two defect prediction frames corresponding to the area index is filtered out.

[0056] Considering that many defect prediction boxes have large overlapping areas in the distribution network scenario, and multiple defect targets with large overlapping areas need to be retained, the filtering threshold is set to a larger value, such as 0.95.

[0057] When the area index exceeds the set filtering threshold, it means that the overlap between the two defect prediction frames is large and filtering is required. In the filtering process, the defect prediction frame with a larger area will be retained, while the defect prediction frame with a smaller area will be filtered out. The area is larger than the defect prediction box The area of ​​the defect prediction box Defect prediction box The area between In the case of , and the defect prediction box Filter out, so that the defect prediction frames with large overlap between adjacent detection frames and small area can be accurately filtered out.

[0058] The preset non-maximum suppression algorithm in this embodiment filters out the defect prediction frames based on their area sizes, sorts them according to their area sizes, gives priority to defect prediction frames with larger areas, and then measures the degree of overlap between the defect prediction frames based on the area index, and sets a higher filtering threshold for filtering, emphasizing the influence of area size on overlapping detection, accurately retaining target frames with larger areas and larger overlapping areas in adjacent detection frames, optimizing the detection results of the target model, and improving the detection accuracy of the target model.

[0059] In one embodiment, a process of filtering defect prediction boxes based on prior experience is described in detail.

[0060] The target model can realize multi-category reasoning such as distribution network defects-equipment-components-towers, that is, in addition to being able to detect component defects in distribution network line images, the target model can also identify the categories of towers and components and generate multiple category prediction boxes.

[0061] According to the prior knowledge of the distribution network scenario, it can be determined that component defect recognition is mainly to identify the defects of valid components in the main tower. In addition to the main tower, the distribution network line image also contains many factors that interfere with the recognition of the target model, such as non-tower information and discarded components. When the target model is disturbed, it will generate defect prediction boxes for invalid defects that are not within the defect recognition range. Based on this, it is necessary to use prior knowledge to filter out the defect prediction boxes belonging to invalid defects in the defect prediction box, and retain the defect prediction boxes for valid components in the main tower.

[0062] In practical applications, the prior knowledge of the distribution network scenario determines that effective defects are distributed on effective components in the main tower. Therefore, the area outside the effective components in the main tower belongs to the distribution area of ​​invalid defects. The area where invalid defects are distributed can be divided into: (1) non-main tower category and (2) waste component category. That is, defects distributed on non-main towers are invalid defects, and defects distributed on waste components are also invalid defects.

[0063] Based on this, the target model can generate multiple category prediction boxes by performing category detection on the distribution network line image. These category prediction boxes can indicate the area where invalid defects are distributed in the distribution network line image, and further filter the multiple defect prediction boxes generated by the target model according to the category prediction boxes to filter out invalid defects in the multiple defect prediction boxes.

[0064] In one embodiment, the category prediction boxes include non-main tower category prediction boxes for indicating non-main towers in the distribution network line image; the category prediction boxes also include discarded component category prediction boxes for indicating discarded components in the distribution network line image.

[0065] In one embodiment, according to the prior knowledge of the network configuration scenario, corresponding filtering rules can be formulated. The specific filtering rules are: (1) When the target detection model infers the non-main tower category prediction box on the distribution network line image, it is necessary to filter out the defect prediction box located in the non-main tower category prediction box.

[0066] It can be understood that the defect prediction box located in the non-main tower category prediction box is the component defect inferred by the target model on the non-main tower, which is an invalid defect. It is necessary to filter out the defects on these non-main towers. (2) When the target detection model infers the discarded component category prediction box on the distribution network line image, it is necessary to calculate whether the discarded component category prediction box intersects with the defect prediction box, and filter out the defect prediction box that intersects with the discarded component category prediction box.

[0067] Among them, the overlapping area between the discarded component category prediction box and the defect prediction box can be calculated, or the intersection-over-union ratio between the discarded component category prediction box and the defect prediction box can be calculated to determine whether the discarded component category and the defect prediction box intersect. If the discarded component category prediction box intersects with the prediction box, it indicates that the target model may have mistakenly identified the features or noise on the discarded components as defects. By filtering out such defect prediction boxes, it is possible to avoid identifying areas with invalid defect distribution as defects, thereby filtering out the misdetected targets of the target model and improving the accuracy of the results.

[0068] In the above embodiment, by introducing a slice reasoning mechanism, the distribution network line image is sliced ​​based on the target model, and the target model is then reasoned based on multiple slice images and distribution network line images to generate multiple defect prediction frames, thereby increasing the target detection model's ability to recognize small targets and improving the recognition accuracy of small targets. Furthermore, a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame is used to filter multiple defect prediction frames, which can filter out prediction frames with large overlapping areas in the distribution network scene, and then multiple defect prediction frames are processed through prior knowledge of the distribution network scene, which can filter out misdetected targets and further improve the recognition accuracy of the target detection model.

[0069] In one embodiment, a device for identifying component defects of a distribution network line is provided, referring to Figure 5 As shown, the component defect identification device 500 for the distribution network line may include: an input module 501, a reasoning module 502 and a filtering module 503. Among them, The input module 501 is used to obtain the distribution network line image and input the distribution network line image into the target model; The reasoning module 502 is used to slice the distribution network line image based on the target model, and generate a prediction box according to multiple slice images and the distribution network line image; wherein the prediction box includes multiple defect prediction boxes; The filtering module 503 is used to adopt a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge to filter multiple defect prediction frames to obtain multiple defect target frames.

[0070] In one embodiment, the filtering module 503 is specifically used to sort the multiple defect prediction boxes in descending order according to the area sizes of the multiple defect prediction boxes; According to the sorted order, the area index between any two defect prediction frames is calculated in sequence; the area index represents the ratio between the overlapping area of ​​the two defect prediction frames and the minimum area of ​​the two defect prediction frames; If the area index is greater than the set filtering threshold, the defect prediction box with the smaller area among the two defect prediction boxes corresponding to the area index will be filtered out.

[0071] In one embodiment, the prediction box also includes multiple category prediction boxes, and the filtering module 503 is specifically used to filter the multiple defect prediction boxes according to the category prediction box; wherein the category prediction box is used to characterize the area of ​​invalid defect distribution determined based on prior knowledge.

[0072] In one embodiment, the category prediction box includes a non-main tower category prediction box and a discarded component category prediction box.

[0073] In one embodiment, the filtering module 503 is specifically configured to, if the category prediction box is a non-main tower category prediction box, filter out the defect prediction box located in the non-main tower category prediction box; If the category prediction box is a discarded component category prediction box, it is calculated whether the discarded component category prediction box intersects with the defect prediction box, and the defect prediction box intersecting with the discarded component category prediction box is filtered out.

[0074] In one embodiment, the reasoning module 502 is specifically used to set cutting parameters, and slice the distribution network line image based on the cutting parameters to obtain multiple slice images; wherein the cutting parameters include: a width direction cutting ratio, a height direction cutting ratio, and a height and width overlapping area ratio; Multiple slice images and distribution network line images are stitched together into a set of data samples.

[0075] In one embodiment, the reasoning module 503 is specifically configured to restore the prediction boxes in the plurality of slice images to the coordinates of the distribution network line image according to the position correspondence between the plurality of slice images and the distribution network line image.

[0076] The specific definition of the component defect identification device 500 for distribution network lines can be found in the above definition of the component defect identification method for distribution network lines, which will not be repeated here. Each module in the above-mentioned component defect identification device 500 for distribution network lines can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0077] In one embodiment, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a method for identifying component defects of a distribution network line is implemented.

[0078] In one embodiment, a computer storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a method for identifying component defects of a distribution network line is implemented.

[0079] It should be noted that the logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, device or equipment (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or equipment and execute instructions), or in combination with these instruction execution systems, devices or equipment. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or equipment, or in combination with these instruction execution systems, devices or equipment. More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection portion with one or more wirings (electronic device), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways if necessary, and then stored in a computer memory.

[0080] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0081] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0082] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0083] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for identifying component defects in a distribution network line, characterized in that: include: Acquire a distribution network line image, and input the distribution network line image into a target model; Slicing the distribution network line image based on the target model, and generating a prediction frame according to a plurality of slice images and the distribution network line image; The prediction box includes a plurality of defect prediction boxes; A preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge are used to filter out multiple defect target frames from the multiple defect prediction frames.

2. The method for identifying component defects of a distribution network line according to claim 1, characterized in that: The preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame filters multiple defect target frames in the multiple defect prediction frames, including: According to the area sizes of the multiple defect prediction boxes, the multiple defect prediction boxes are sorted in descending order; According to the sorted order, the area index between any two of the defect prediction frames is calculated in sequence; the area index represents the ratio between the overlapping area of ​​the two defect prediction frames and the minimum area of ​​the two defect prediction frames; If the area index is greater than a set filtering threshold, the defect prediction frame with a smaller area among the two defect prediction frames corresponding to the area index is filtered out.

3. The method for identifying component defects of distribution network lines according to claim 2, characterized in that: The prediction frame further includes a plurality of category prediction frames, and the preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge are used to filter out a plurality of defect target frames from the plurality of defect prediction frames, including: The plurality of defect prediction frames are filtered according to the category prediction frame; the category prediction frame is used to characterize the region of invalid defect distribution determined based on the prior knowledge.

4. The method for identifying component defects of distribution network lines according to claim 3, characterized in that: The category prediction box includes a non-main tower category prediction box and a discarded component category prediction box.

5. The method for identifying component defects of distribution network lines according to claim 4, characterized in that: The filtering process of the plurality of defect prediction frames according to the category prediction frame includes: If the category prediction frame is a non-main tower category prediction frame, filtering out the defect prediction frame located in the non-main tower category prediction frame; If the category prediction box is a discarded component category prediction box, it is calculated whether the discarded component category prediction box intersects with the defect prediction box, and the defect prediction box intersecting with the discarded component category prediction box is filtered out.

6. The method for identifying component defects of distribution network lines according to claim 1, characterized in that: The slicing process of the distribution network line image based on the target model and generating a prediction frame according to a plurality of slice images and the distribution network line image includes: Setting cutting parameters, and slicing the distribution network line image based on the cutting parameters to obtain a plurality of sliced ​​images; the cutting parameters include: a width direction cutting ratio, a height direction cutting ratio, and a height and width overlapping area ratio; A plurality of the slice images and the distribution network line images are spliced ​​into a set of data samples.

7. The method for identifying component defects of distribution network lines according to claim 1, characterized in that: The generating a prediction frame according to the plurality of slice images and the distribution network line image comprises: According to the position correspondence between the plurality of slice images and the distribution network line image, the prediction boxes in the plurality of slice images are restored to the coordinates of the distribution network line image.

8. A device for identifying component defects in a distribution network line, characterized in that: include: An acquisition module, used for acquiring a distribution network line image and inputting the distribution network line image into a target model; An inference module, configured to slice the distribution network line image based on the target model, and generate a prediction frame according to a plurality of slice images and the distribution network line image; The prediction box includes a plurality of defect prediction boxes; The filtering module is used to adopt a preset non-maximum suppression algorithm based on the area of ​​the defect prediction frame and prior knowledge to filter multiple defect target frames from the multiple defect prediction frames.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for identifying component defects in distribution network lines described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying component defects in distribution network lines described in any one of claims 1 to 7 are implemented.

Images