Power transmission line fault detection method and system based on lightweight target detection model

Abstract

The application provides a power transmission line fault detection method and system based on a lightweight target detection model, which comprises a power transmission line image shooting step, a power transmission line image processing step, a power transmission line inspection image detection step and a power transmission line defect identification and classification step. The power transmission line image shooting step is to shoot a power transmission line image to be detected, the power transmission line image processing step is to process the shot power transmission line image, so that the image can be recognized by a lightweight target detection network, the power transmission line inspection image detection step is to detect whether the power transmission line image to be detected has defects and damages according to a trained model, and the power transmission line defect identification and classification step is to classify and spatially locate the defects and damages if any after the power transmission line inspection image detection step. The method has the advantages that a modified lightweight target detection model is used, and the non-contact and real-time detection effect with low calculation cost and consumption can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of target detection in computer vision, and in particular to a power transmission line fault detection method and system based on a lightweight target detection model. BACKGROUND

[0002] Target detection is a challenging field in computer vision, and its results have been widely applied to many fields. Fault detection of power transmission lines mostly uses satellite imaging or unmanned aerial vehicle line inspection. The latter generally uses unmanned aerial vehicles to take images of fixed lines, and then transmits them to the monitoring center for fault detection by target detection methods or manpower. This method requires high human and financial resources, and does not have real-time detection and classification effects.

[0003] The YOLO series algorithm has been a mature target detection method after years of updates and iterations. In the latest YOLOv7 target detection algorithm, the speed and accuracy exceed all known target detectors in most cases. However, the original YOLOv7 has added a lot of other methods through the development of the YOLO series, becoming a mature and complex convolutional neural network. This network can run excellent results on high-computing devices, but is not suitable for real-time fault detection work in unmanned aerial vehicles for power line inspection. SUMMARY

[0004] The purpose of the present application is to provide a power transmission line fault detection method and system based on a lightweight target detection model, to solve or at least partially solve the technical problem that real-time and classification effects cannot be considered in the prior art.

[0005] In order to solve the above technical problems, the present application adopts the following technical solutions:

[0006] The first aspect provides a power transmission line fault detection method based on a lightweight target detection model, comprising:

[0007] Collecting power transmission line images;

[0008] Preprocessing the collected power transmission line images;

[0009] The pre-processed power transmission line image is positioned for defects by using a pre-constructed lightweight target detection model, wherein the lightweight target detection model is an improved model LC-YOLOv7 based on YOLOv7, the model includes a backbone feature extraction network, a strengthened feature extraction network and a detection head part, the improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a group of MP1 and ELAN structures to the YOLOv7 model to obtain more resolution feature maps, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the added feature maps, and the detection head part adds a detection head to the YOLOv7 model to detect different sizes of power transmission line faults, wherein the MP1 structure is a special convolution combination including three convolution layers and a maximum pooling layer, the output channel number of the MP1 is the same as the input channel number, the ELAN structure is a convolution fusion structure used for fusing feature information obtained by multiple convolution layers, the ELAN structure includes six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the sixth convolution layer controls the channel number of the output;

[0010] According to the defect positioning result of the power transmission line image, the pre-constructed lightweight target detection model is used for defect classification and spatial positioning.

[0011] In an embodiment, the method further comprises:

[0012] According to the defect positioning result, the defect classification and the spatial positioning result, an injury alarm is given.

[0013] In an embodiment, the power transmission line image is collected, including:

[0014] According to the pre-prepared power transmission line inspection line, a UAV deployed with the LC-YOLOv7 model is used to collect the power transmission line image along the line.

[0015] In an embodiment, the collected power transmission line image is pre-processed, including:

[0016] The power transmission line image is subjected to resolution normalization processing; the shooting order of the to-be-detected power transmission line image is matched with the power transmission line inspection order, the power transmission line image is uniformly converted into PNG or JPG format and the resolution of the image is made to reach a preset range.

[0017] In an embodiment, the pre-processed power transmission line image is positioned for defects by using a pre-constructed lightweight target detection model, including:

[0018] Four feature maps with different resolutions are obtained from the input pre-processed power transmission line image by the backbone feature extraction network;

[0019] The two feature pyramids of the enhanced feature extraction network gradually perform feature fusion on four feature maps with different resolutions;

[0020] The feature maps after feature fusion are input into the four detection heads one by one, and the coordinates of defects of different sizes in the transmission line image are determined through anchor boxes of different sizes in the detection heads, as the defect positioning results.

[0021] In an embodiment, when there is a defect loss in the preprocessed transmission line image, the method further comprises:

[0022] Repeatedly shooting the position of the defect;

[0023] Inputting the image obtained by repeated shooting into the lightweight target detection model for re-detection;

[0024] Artificially detecting the repeated shooting image.

[0025] In an embodiment, according to the defect positioning result of the transmission line image, a pre-constructed lightweight target detection model is used for defect classification and spatial positioning, comprising:

[0026] Classifying the defect type according to the defect positioning result;

[0027] Determining the spatial position of the defect according to the transmission line inspection line and the defect positioning result.

[0028] Based on the same inventive concept, the second aspect of the present application provides a transmission line fault detection system based on a lightweight target detection model, comprising:

[0029] A transmission line image acquisition module for acquiring a transmission line image;

[0030] A transmission line image preprocessing module for preprocessing the acquired transmission line image;

[0031] The power transmission line image detection module is used for defect positioning on the preprocessed power transmission line image by using a pre-constructed lightweight target detection model, wherein the lightweight target detection model is an improved model LC-YOLOv7 based on YOLOv7, the model comprises a backbone feature extraction network, a strengthened feature extraction network and a detection head part, the improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a group of MP1 and ELAN structures on the basis of the YOLOv7 model to obtain more resolution feature maps, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the added feature maps, and the detection head part adds a detection head on the basis of the YOLOv7 model to detect power transmission line faults of different sizes, wherein the MP1 structure is a special convolution combination, comprising three convolution layers and a maximum pooling layer, the output channel number of the MP1 is the same as the input channel number, the ELAN structure is a convolution fusion structure and is used for fusing feature information obtained by multiple convolution layers, the ELAN structure comprises six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the sixth convolution layer is used for controlling the channel number of the output;

[0032] The power transmission line defect recognition and classification module is used for defect classification and spatial positioning by using the pre-constructed lightweight target detection model according to the defect positioning result of the power transmission line image

[0033] Based on the same inventive concept, the third aspect of the present application provides a computer readable storage medium, which stores a computer program, and the program is executed to realize the method of the first aspect.

[0034] Based on the same inventive concept, the fourth aspect of the present application provides a computer device, which comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor executes the program to realize the method of the first aspect.

[0035] Compared with the prior art, the present application has the following advantages and beneficial technical effects:

[0036] The power transmission line fault detection method based on the lightweight target detection model disclosed in the application uses an improved model LC-YOLOv7 based on YOLOv7 to perform defect positioning and classification identification, so as to realize fault inspection. The improved model reduces the number of channels of the overall network, thereby reducing the number of convolution calculations, making the network more lightweight, then adding a group of convolutions in the backbone feature extraction network to obtain feature maps of different resolutions, and increasing the three-layer double feature pyramid in the enhanced feature extraction network to four layers to fuse the newly added feature maps, finally inputting the group of feature maps into a newly added detection head, so that the LC-YOLOv7 network can detect more different sizes of faults than the original YOLOv7, thereby ensuring the target detection accuracy after lightening. The LC-YOLOv7 network improved in this way can ensure a higher detection accuracy and meet the real-time monitoring requirements of power transmission line inspection. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0038] Figure 1 is a brief flowchart of the power transmission line fault detection method based on the lightweight target detection model provided by the embodiment of the application;

[0039] Figure 2 is a network structure diagram of the lightweight target detection model used in the fault detection method in the embodiment of the application;

[0040] Figure 3 is a schematic diagram of the change of feature maps and channel numbers in the power transmission line fault detection method based on the lightweight target detection model in the embodiment of the application;

[0041] Figure 4 is an actual application flowchart of the power transmission line fault detection method based on the lightweight target detection model provided by the embodiment of the application. DETAILED DESCRIPTION

[0042] Since the complexity of the YOLO series target detection algorithm has been greatly improved, it is not suitable for the low cost and real-time requirements of power transmission line inspection. Therefore, in order to meet the real-time monitoring requirements of power transmission line inspection under the premise of a higher detection accuracy, the application proposes a power transmission line fault detection method based on a lightweight target detection model.

[0043] The main advantage and value of the present application are that the LC-YOLOv7 improved based on the YOLOv7 target detection algorithm is used for power line inspection, the YOLOv7 has quite high target detection accuracy compared with other target detection algorithms, and the LC-YOLOv7 is improved in light weight under the premise of maintaining the detection accuracy of the YOLOv7, so that the LC-YOLOv7 can be carried on the low-computing-capability unmanned aerial vehicle for power line inspection, and meet the real-time detection requirement, so as to timely detect the defect position and classify the defect category of the power line, and achieve the effect of saving manpower and material resources.

[0044] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] Embodiment one

[0046] The embodiments of the present application provide a power line fault detection method based on a light-weight target detection model, which comprises the following steps:

[0047] Collecting power line images;

[0048] Pretreating the collected power line images;

[0049] Positioning defects in the pretreated power line images by using a pre-constructed light-weight target detection model, wherein the light-weight target detection model is an improved model LC-YOLOv7 based on YOLOv7, the model comprises a backbone feature extraction network, a strengthened feature extraction network and a detection head part, the improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a group of MP1 and ELAN structures to obtain more resolution feature maps on the basis of the YOLOv7 model, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the added feature maps, and the detection head part adds a detection head on the basis of the YOLOv7 model to detect power line faults of different sizes, wherein the MP1 structure is a special convolution combination comprising three convolution layers and a maximum pooling layer, the output channel number of the MP1 is the same as the input channel number, the ELAN structure is a convolution fusion structure used for fusing feature information obtained by multiple convolution layers, the ELAN structure comprises six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the channel number of the output is controlled by the sixth convolution layer;

[0050] According to the defect positioning result of the power transmission line image, a pre-constructed lightweight target detection model is used for defect classification and spatial positioning.

[0051] See Figure 1 , a brief flowchart of the power transmission line fault detection method based on the lightweight target detection model provided by the embodiment of the present application; the method of the embodiment of the present application includes power transmission line image shooting (collection), power transmission line image processing, power transmission line inspection image detection, power transmission line defect identification and classification steps, and power transmission line damage alarm steps (optional) as a whole.

[0052] In the specific implementation process, the power transmission line image shooting step can be realized by shooting the power transmission line image to be detected according to the predetermined power transmission line inspection route by the unmanned aerial vehicle; the power transmission line image processing step is data processing of the shot power transmission line image, which ensures that the image can be recognized and processed by the lightweight target detection network LC-YOLOv7; the power transmission line inspection image detection step is to detect whether there is a defect and damage in the power transmission line image to be detected according to the LC-YOLOv7 model trained by the previous power transmission line image and to detect the defect and damage after a certain detection accuracy is reached; the power transmission line defect identification and classification step is to classify the defect or damage by LC-YOLOv7 to determine the repair method if a defect or damage is found after the power transmission line inspection image detection step. Optionally, the power transmission line damage alarm step can be included, different alarms can be sent according to the power transmission line damage position detected by the power transmission line inspection image detection step and the power transmission line damage category classified by the power transmission line defect identification and classification step, so as to ensure that the power transmission line damage is handled. The method of the present application is to identify the power transmission line damage and defect based on the target detection technology, and the advantage lies in that a modified lightweight target detection model is used, which can be carried on the unmanned aerial vehicle with poor computing ability, realizing the non-contact and real-time detection effect with low computing cost and consumption.

[0053] In one embodiment, the method further comprises:

[0054] According to the defect positioning result, the defect classification and spatial positioning result, the damage alarm is performed.

[0055] In the specific implementation process, the defect and damage information encountered by the unmanned aerial vehicle in the power transmission line inspection, including the position, category and repeated shooting image, can be submitted to the manual monitoring center and an alarm is sent according to the power transmission line damage found in the power transmission line inspection image detection step and the power transmission line damage type and spatial position confirmed in the power transmission line defect identification and classification step.

[0056] In one embodiment, the power transmission line image is collected, comprising:

[0057] According to the pre-prepared power transmission line inspection line, the unmanned aerial vehicle deployed with the LC-YOLOv7 model is used to collect power transmission line images along the line.

[0058] In an embodiment, the collected power transmission line images are pre-processed, including:

[0059] The power transmission line images are subjected to resolution normalization processing; the shooting order of the to-be-detected power transmission line images is matched with the power transmission line inspection order, the power transmission line images are uniformly converted into PNG or JPG format and the resolution of the images is made to reach a preset range.

[0060] In the specific implementation process, the resolution normalization processing is performed on the power transmission line images to ensure the stability of the subsequent input detection step. For the previous power transmission line images used to train the LC-YOLOv7 model, the training set, the validation set and the test set should be divided according to the ratio of 8:1:1. The shooting order of the to-be-detected power transmission line images should be matched with the power transmission line inspection order so as to facilitate the positioning and repeated shooting when the defect loss is found. The power transmission line images should be uniformly converted into PNG or JPG format and the images that cannot be distinguished or have poor quality should be re-shot.

[0061] In an embodiment, a pre-constructed lightweight target detection model is used to locate defects in the pre-processed power transmission line images, including:

[0062] Four feature maps with different resolutions are obtained from the input pre-processed power transmission line images through a backbone feature extraction network;

[0063] The four feature maps with different resolutions are gradually subjected to feature fusion through two feature pyramids of the enhanced feature extraction network;

[0064] The feature maps subjected to feature fusion are input into four detection heads one by one, and the coordinate positions of defects and damages with different sizes in the power transmission line images are determined through anchor boxes with different sizes in the detection heads as the defect positioning results.

[0065] The present application proposes to replace the previous complex YOLOv7 target detection network with the LC-YOLOv7 lightweight target detection network, and the network structure of the LC-YOLOv7 is as shown in Figure 2 Figure 2 The left side is the network structure of the LC-YOLOv7, and the right side is the specific module structure diagram, Figure 2 The range marked by the thick dashed line frame in the left network structure diagram is the newly added module of the model of the present application. The following improvements are made in the specific network model:

[0066] ​One: reduce the number of channels of the whole network to 20%. The 640*640*3 in the figure means that the feature map of the current module is 640*640 resolution and 3 channels. The channel number in the figure is the data after it has been reduced to 20%.

[0067] Two: add a set of MP1 and ELAN structures to the backbone feature network to obtain more resolution feature maps. The backbone feature extraction network can obtain 10*10 resolution feature maps, while the original YOLOv7 can only obtain 20*20 resolution.

[0068] Three: strengthen the feature extraction network by changing the double feature pyramid from 3 layers to 4 layers to fuse the newly added 10*10 resolution feature maps.

[0069] Four: add a detection head corresponding to the new resolution feature map, so that the whole network model can detect more different sizes of power line faults.

[0070] Specifically, the backbone feature extraction network includes multiple groups of standard convolution, MP convolution and ELAN structure, which can obtain four kinds of feature maps with different resolutions, and then input into the strengthened feature extraction network.

[0071] The CBS in the LC-YOLOv7 is a standard convolution module, including convolution, normalization layer and activation function. The MP model includes two types of MP1 and MP2, both of which are a special convolution combination, including three convolutions and a max pooling layer, the difference is that the output channel number of MP1 is the same as the input channel number, while the output channel number of MP2 is twice the input channel number, the purpose is to fuse multiple semantic information in the feature map while convolution. The ELAN structure and the ELAN-H structure also fuse the feature information obtained by multiple convolutions, including six convolutions, the first five convolutions are fused to obtain different feature maps, and the last convolution controls the number of output channels, the difference between ELAN and ELAN-H structure is the fusion method of the first five convolutions, if the input channel number is C, the ELAN structure fuses four convolutions with C / 2, and the ELAN-H structure fuses two convolutions with C / 2 and the remaining four convolutions with C / 4. The SPPCSPC structure includes seven convolutions and three max pooling layers, the main purpose is to obtain multi-dimensional feature information through three max pooling layers with different pooling kernels for the smallest resolution and the most semantic features in the backbone feature extraction network, so as to further fuse the feature extraction network. RepConv is a planned reparameterization convolution, which is a standard residual structure convolution combination when training the model, but becomes a degenerate 3*3 convolution through reparameterization when verifying and detecting power line defects, which is beneficial to actual testing and deployment. Other includes UP structure, which is a fixed combination of a convolution that controls the number of channels and upsampling. And the Concat function is used to add multiple input feature maps.

[0072] See Figure 3 , which is a feature map and channel number change schematic diagram in the power line fault detection method based on the lightweight target detection model in the embodiment of the application.

[0073] Specifically, the enhanced feature extraction network contains two feature pyramids, each of which includes four layers, which gradually fuse the features of different resolutions obtained in the previous step to ensure that the LC-YOLOv7 network pays attention to various sizes of power line defect damage. Then input the feature maps after feature fusion into the detection head one by one, and determine the coordinate position of defects of different sizes in the power line image by the anchor boxes of different sizes in the detection head.

[0074] In one embodiment, when there is a defect loss in the preprocessed power line image, the method further comprises:

[0075] Repeatedly shooting the position of the defect;

[0076] Inputting the images obtained by repeated shooting into the lightweight target detection model for re-detection;

[0077] The repeated image is subjected to manual defect detection.

[0078] Specifically, the LC-YOLOv7 model trained in the power line image processing step should have a detection accuracy of more than 85%, and if the detection accuracy is not reached, the training set should be increased or the image shooting method should be improved to improve the detection accuracy.

[0079] If a defect loss is found in the image, the power line image shooting step is returned to the location of the defect to take repeated images, first re-detect the repeated images to ensure detection accuracy, and then submit the repeated images for manual confirmation of whether there is a power line defect loss. Please refer to Figure 4 , is the actual application flowchart of the power line fault detection method based on the lightweight target detection model provided by the embodiment of the application.

[0080] In one embodiment, according to the defect positioning result of the power line image, a pre-constructed lightweight target detection model is used for defect classification and spatial positioning, including:

[0081] Classify the defect type according to the defect positioning result;

[0082] According to the power line inspection line and the defect positioning result, the spatial position of the defect is determined.

[0083] In the specific implementation process, the power line defects found in the power line inspection image detection step are classified, including external damage, acid and alkali corrosion, power line shedding, etc. The damage type is determined and the real spatial position of the damage is determined according to the image shooting order of the circuit inspection line and the coordinate position of the defect damage in the power line image in the power line inspection image detection step.

[0084] Overall, the advantages and beneficial technical effects of the present application are as follows:

[0085] LC-YOLOv7 is used as the target detection algorithm for power line inspection, and LC-YOLOv7 is improved on the latest YOLO series algorithm YOLOv7. First, the channel number is reduced to greatly lighten the overall network. However, this will cause the detection accuracy to decrease, so LC-YOLOv7 adds a group of convolution and its corresponding module to detect more different sizes of power line faults, thereby ensuring that the overall detection accuracy changes little. The improved LC-YOLOv7 is more easily deployed on the unmanned aerial vehicle for power line inspection and achieves the effect of real-time detection, which can quickly find power line fault defects and classify defect types during the inspection process, thereby issuing an alarm to reduce possible harm.

[0086] Example Two

[0087] Based on the same inventive concept, the application discloses a power transmission line fault detection system based on a lightweight target detection model, comprising:

[0088] A power transmission line image acquisition module is configured to acquire power transmission line images.

[0089] A power transmission line image preprocessing module is configured to preprocess the acquired power transmission line images.

[0090] A power transmission line image detection module is configured to locate defects in the preprocessed power transmission line images using a pre-constructed lightweight target detection model, wherein the lightweight target detection model is an improved model LC-YOLOv7 based on YOLOv7, and the model includes a backbone feature extraction network, a strengthened feature extraction network, and a detection head part. The improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a set of MP1 and ELAN structures to the YOLOv7 model to obtain more resolution feature maps, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the newly added feature maps, and the detection head part adds a detection head to the YOLOv7 model to detect different sizes of power transmission line faults. The MP1 structure is a special convolution combination, including three convolution layers and a maximum pooling layer, and the output channel number of the MP1 is the same as the input channel number. The ELAN structure is a convolution fusion structure used to fuse the feature information obtained by multiple convolution layers. The ELAN structure includes six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the channel number of the output is controlled by the sixth convolution layer.

[0091] A power transmission line defect recognition and classification module is configured to classify and spatially locate defects using the pre-constructed lightweight target detection model based on the defect location results of the power transmission line images.

[0092] Since the system introduced in Example Two of the application is used to implement the power transmission line fault detection method based on the lightweight target detection model in Example One of the application, the specific structure and variations of the system can be understood by those skilled in the art based on the method introduced in Example One of the application, and therefore will not be described here. Any system used in the method of Example One of the application belongs to the scope of protection of the present application.

[0093] Example Three

[0094] Based on the same inventive concept, the application further provides a computer readable storage medium having a computer program stored thereon, the program being executed to implement the method as described in Example One.

[0095] Since the computer readable storage medium introduced in the embodiment three of the present application is the computer readable storage medium used for implementing the power line fault detection method based on the lightweight target detection model in the embodiment one of the present application, the specific structure and deformation of the computer readable storage medium can be understood by the person skilled in the art based on the method introduced in the embodiment one of the present application, and thus will not be described here again. The computer readable storage medium used by the method in the embodiment one of the present application belongs to the scope of the present application.

[0096] Embodiment four

[0097] Based on the same inventive concept, the present application further provides a computer device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, and the processor implements the method in the embodiment one when executing the program.

[0098] Since the computer device introduced in the embodiment four of the present application is the computer device used for implementing the power line fault detection method based on the lightweight target detection model in the embodiment one of the present application, the specific structure and deformation of the computer device can be understood by the person skilled in the art based on the method introduced in the embodiment one of the present application, and thus will not be described here again. The computer device used by the method in the embodiment one of the present application belongs to the scope of the present application.

[0099] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0100] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The device for implementing the function specified in one block or multiple blocks.

[0101] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0102] It is apparent that those skilled in the art can, without departing from the spirit and scope of the application, make various changes and modifications to the embodiments of the application. As such, it is intended that the application cover all such changes and modifications that fall within the scope of the application.

Claims

1. A power transmission line fault detection method based on a lightweight target detection model, characterized in that, The method comprises the following steps: collecting power transmission line images; preprocessing the collected power transmission line images; using a pre-constructed lightweight target detection model to locate defects in the preprocessed power transmission line images, wherein the lightweight target detection model is an improved model LC-YOLOv7 based on YOLOv7, the model includes a backbone feature extraction network, a strengthened feature extraction network and a detection head part, the improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a set of MP1 and ELAN structures to the YOLOv7 model to obtain more resolution feature maps, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the added feature maps, and the detection head part adds a detection head to the YOLOv7 model to detect different sizes of power transmission line faults, wherein the MP1 structure is a special convolution combination including three convolution layers and a maximum pooling layer, the output channel number of the MP1 is the same as the input channel number, the ELAN structure is a convolution fusion structure used for fusing feature information obtained by multiple convolution layers, and the ELAN structure includes six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the sixth convolution layer controls the channel number of the output; according to the defect location result of the power transmission line image, using the pre-constructed lightweight target detection model to classify and locate the defects in space. 2.The power transmission line fault detection method based on the lightweight target detection model of claim 1, wherein, The method further comprises: issuing a damage warning according to the defect location result, the defect classification and the spatial location result. 3.The power transmission line fault detection method based on the lightweight target detection model of claim 1, wherein, The method further comprises: collecting power transmission line images, including: 4.The power transmission line fault detection method based on the lightweight target detection model of claim 1, wherein, according to a pre-prepared power transmission line inspection route, using a drone deployed with the LC-YOLOv7 model to collect power transmission line images along the route. preprocessing the collected power transmission line images, including: 5.The power transmission line fault detection method based on the lightweight target detection model of claim 1, wherein, performing resolution normalization on the power transmission line images; matching the shooting order of the to-be-detected power transmission line images with the power transmission line inspection order, converting the power transmission line images into PNG or JPG format, and ensuring that the resolution of the images reaches a preset range. using a pre-constructed lightweight target detection model to locate defects in the preprocessed power transmission line images, including: obtaining feature maps with different resolutions from the input preprocessed power transmission line images through the backbone feature extraction network; gradually fusing the feature maps with different resolutions through the two feature pyramids of the strengthened feature extraction network; 6.The power transmission line fault detection method based on the lightweight target detection model according to claim 5, wherein, inputting the feature maps after feature fusion into the four detection heads one by one, determining the coordinate positions of defects with different sizes in the power transmission line images through anchor boxes with different sizes in the detection heads, and taking the coordinate positions as the defect location result. When there are defects in the preprocessed power transmission line images, the method further comprises: repeatedly shooting the positions of the defects; inputting the images obtained by repeated shooting into the lightweight target detection model for re-detection; 7.The power transmission line fault detection method based on the lightweight target detection model of claim 1, wherein, performing manual defect detection on the repeatedly shot images. According to the defect location result of the power transmission line image, using the pre-constructed lightweight target detection model to classify and locate the defects in space, including: According to the defect positioning result, the defect type is classified; According to the transmission line inspection line and the defect positioning result, the spatial position of the defect is determined.

8. A power transmission line fault detection system based on a lightweight object detection model, characterized by, The method comprises the steps of: The transmission line image acquisition module is configured to acquire a transmission line image. The transmission line image preprocessing module is configured to preprocess the acquired transmission line image. The transmission line image detection module is configured to perform defect positioning on the preprocessed transmission line image by using a pre-constructed lightweight target detection model, wherein the lightweight target detection model is an improved model LC-YOLOv7 based on YOLOv7, the model comprises a backbone feature extraction network, a strengthened feature extraction network and a detection head part, the improved model LC-YOLOv7 reduces the channel number of the YOLOv7 model, the backbone feature extraction network adds a set of MP1 and ELAN structures to the YOLOv7 model to obtain more resolution feature maps, the strengthened feature extraction network changes the double feature pyramid of the YOLOv7 model from 3 layers to 4 layers to fuse the newly added feature maps, and the detection head part adds a detection head to the YOLOv7 model to detect different sizes of transmission line faults, wherein the MP1 structure is a special convolution combination comprising three convolution layers and a maximum pooling layer, the output channel number of the MP1 is the same as the input channel number, the ELAN structure is a convolution fusion structure for fusing feature information obtained by multiple convolution layers, the ELAN structure comprises six convolution layers, different feature maps obtained by the first five convolution layers are fused, and the sixth convolution layer controls the channel number of the output; The transmission line defect recognition and classification module is configured to perform defect classification and spatial positioning by using the pre-constructed lightweight target detection model according to the defect positioning result of the transmission line image.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed to implement the method of any one of claims 1 to 7.

10. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the method of any one of claims 1 to 7.

Images