Power cable surface defect detection method, device, equipment and medium

Abstract

The invention belongs to the technical field of image processing, and relates to a power cable surface defect detection method, device, equipment and medium, and the method comprises the following steps: obtaining and preprocessing cable sample images with different surface defects; marking the cable sample image to identify the position and the type of a defect; training a cable surface defect detection model based on the SSD network by using the marked cable sample image, and extracting and classifying defect features by using a ResNet50 network as a backbone network to replace a VGG16 network in the original SSD network; and inputting a to-be-detected image into the trained cable surface defect detection model, and automatically detecting defects on the surface of the cable. Compared with the prior art, the method not only can improve the accuracy and efficiency of power cable surface defect detection, but also can enhance the generalization ability and stability of the model.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, equipment and medium for detecting surface defects of power cables based on deep learning. Background Art

[0002] In the field of power cable surface defect detection, with the acceleration of urbanization and the increase in electricity demand, the reliability and safety requirements of power cables are getting higher and higher. Traditional manual detection methods are inefficient and easily affected by subjective judgment. Therefore, power cable surface defect detection methods based on deep learning have become a hot topic of research. Deep learning technology has shown excellent performance in image recognition and processing. By training models to identify defects on the cable surface, such as cracks, depressions, and protrusions, the efficiency and accuracy of detection can be greatly improved. For example, by collecting image data on the cable surface, and then using deep learning algorithms to extract and classify features of the image, it is possible to identify whether there are defects and the type and severity of the defects.

[0003] Patent CN 202410209158.3 proposes a cable accessory defect detection method, device, equipment and storage medium. The method collects on-site images of cable accessories and performs image processing to obtain cable joint images; improves the feature extraction network and regional pooling of the Faster RCNN model to obtain an improved Faster RCNN model; and detects the cable joint image based on the improved Faster RCNN model to obtain the defect type and defect location. The advantage of this invention is that it utilizes the high accuracy of the FasterRCNN model, especially in detecting small targets and multi-scale targets. However, this two-stage detection method also leads to a relatively slow detection speed.

[0004] Patent CN 202110453748.7 proposes a method for detecting cable surface defects by drone intelligent inspection based on deep learning. The YOLOv3 surface defect detection model is used for training and detection respectively. The defect detection model constructed by the drone-collected cable image data verification and application has a good recognition ability for the early stage of defects in the cable, and can detect cable defects in time and deal with them quickly. The advantage of this invention is that it can achieve near real-time detection by taking advantage of the fast speed of the YOLO model, which is very useful for application scenarios that require fast response, but relatively speaking, the accuracy of the YOLO model is not precise enough.

[0005] The SSD model has significant advantages in the application of surface defect detection. The SSD network combines the advantages of Faster-RCNN and YOLO. SSD is a single-stage detection method that can effectively detect targets of different sizes by performing target detection on feature maps of multiple scales, which enables it to maintain good performance when dealing with both small and large targets. This is especially important in cable surface defect detection because defects may appear in many forms and sizes. However, SSD has the problems of weak generalization ability and poor reliability. Summary of the invention

[0006] The purpose of the present invention is to provide a method, device, equipment and medium for power cable surface defect detection based on deep learning. Replacing the VGG16 network in the SSD network with the ResNet50 network can not only improve the accuracy and efficiency of power cable surface defect detection, but also enhance the generalization ability and stability of the model.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A method for detecting surface defects of power cables based on deep learning, comprising the following steps:

[0009] Acquire and preprocess images of cable samples with different surface defects;

[0010] Annotate cable sample images to identify defect locations and types;

[0011] The labeled cable sample images are used to train the cable surface defect detection model based on the SSD network. The ResNet50 network is used as the backbone network to replace the VGG16 network in the original SSD network to extract and classify defect features.

[0012] The image to be inspected is input into the trained cable surface defect detection model to automatically detect defects on the cable surface.

[0013] The use of the ResNet50 network as the backbone network to replace the VGG16 network in the original SSD network is specifically as follows:

[0014] The conv2_x, conv3_x, and conv4_x in the ResNet50 network are extracted as feature layers and fused, and the last fully connected layer in the original SSD network is removed; the fused feature layer is used as the first convolution layer, and then 6 convolution layers with different convolution kernel sizes are added in sequence behind the convolution layer of the ResNet50 network to generate feature maps of different scales, which are used to detect targets of different sizes.

[0015] At each convolutional layer, an anchor box is generated based on the size of the feature map and the size of the expected detection object.

[0016] The steps for generating the anchor point frame are as follows:

[0017] Define the anchor box scale. The anchor box scale calculation formula is as follows:

[0018]

[0019] Among them, S k Indicates the proportion of the anchor box size in the image, S min Indicates the minimum size ratio of the anchor box, S max represents the maximum size ratio of the anchor box, and m represents the number of feature layers;

[0020] Define the anchor box aspect ratio α r , calculate the width and height of the anchor box based on the aspect ratio of the anchor box:

[0021]

[0022] Among them, w k Indicates the width of the anchor box, h k Indicates the height of the anchor box;

[0023] Calculate the center coordinates of each anchor box:

[0024]

[0025] Among them, |f k | is the size of the kth feature map, a,b∈{0,1,2,…,|f k -1|}.

[0026] After the cable surface defect detection model completes the prediction in all feature layers, it selects the anchor box with the highest prediction score from each feature layer, and excludes highly overlapping anchor boxes through the NMS algorithm. The retained anchor boxes serve as the prediction boxes to form the final detection results, which include the location, confidence and category of the target.

[0027] The loss function of the cable surface defect detection model consists of positioning loss and confidence loss, and its formula is:

[0028]

[0029] Among them, L conf (x,c) represents the confidence loss, which is used to calculate the difference between the confidence of the predicted category and the true category; L loc(x, l, g) is the positioning loss, which is used to calculate the distance difference between the predicted box and the true box; N represents the number of positive samples, x represents the matching result between the predicted box and the true box, c represents the category confidence prediction value, l represents the position prediction value of the predicted box, g represents the position parameter of the true box, and α is the weight coefficient.

[0030] The pre-processing comprises the following steps:

[0031] CLAHE image enhancement processing: the cable sample image is divided into small blocks, and histogram equalization is applied locally to enhance the local contrast by limiting the contrast while avoiding enhancing the background noise;

[0032] Image filtering: Use median filtering to filter the image that has undergone image enhancement processing.

[0033] A power cable surface defect detection device based on deep learning comprises a memory, a processor, and a program stored in the memory, wherein the processor implements the method described above when executing the program.

[0034] An electronic device comprises a memory and a processor, wherein a computer program is stored in the memory, and the method described above is implemented when the processor executes the program.

[0035] A storage medium stores a program, which implements the method described above when executed.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) Stability: The present invention adopts the ResNet50 network combined with the SSD network. Compared with the traditional VGG16 network, the ResNet50 solves the gradient vanishing and gradient exploding problems of the deep network through residual learning in design, so it has better generalization ability. In addition, the residual connection of ResNet50 helps to reduce the overfitting phenomenon of the network, making the model more stable and reliable in practical applications.

[0038] (2) Accuracy: The present invention applies the deep learning model to cable surface defect detection, and the deep learning algorithm can automatically learn and extract complex features from image data, thereby achieving high-precision defect detection.

[0039] (3) Flexibility: The present invention applies deep learning models to cable surface defect detection, and deep learning models can be flexibly deployed on various platforms, including servers, workstations, embedded devices and even mobile devices. Therefore, they have a wide range of applications and strong flexibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1is a flow chart of the method of the present invention;

[0041] Figure 2 This is a schematic diagram of a cable surface defect detection model of the present invention;

[0042] Figure 3 A PR curve comparison diagram of cable surface defect detection using different models in an embodiment;

[0043] Figure 4 This is a structural diagram of a power cable surface defect detection device of the present invention;

[0044] Figure 5 This is a structural diagram of an electronic device of the present invention. DETAILED DESCRIPTION

[0045] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0046] Example 1

[0047] This embodiment provides a method for detecting surface defects of power cables based on deep learning. Figure 1 As shown, the following steps are included:

[0048] S1, acquire and preprocess cable sample images with different surface defects.

[0049] In this embodiment, 3788 images containing 3 types of cable surface defects are collected, and a new folder named VOC is created under the PyTorch platform. Three subfolders are created under the VOC folder, with the default names Annotations, lmageSets, and JPEGlmages. The JPEGlmages folder stores image files, the Annotations folder stores label files with the suffix xml, and the lmageSets folder stores data set files such as train.txt, test.txt, val.txt, and trainval.txt. In the data set, the ratio of training set, validation set, and test set is set to 7:2:1. The images need to be preprocessed before training. In this embodiment, the preprocessing includes the following steps:

[0050] S11, CLAHE image enhancement processing: The CLAHE algorithm divides the image into small blocks, applies histogram equalization locally, and then uses the method of limiting contrast to enhance the local contrast while avoiding enhancing background noise. This technology is particularly suitable for images with low contrast in local areas and can significantly improve the visual effect of the image.

[0051] S12, image filtering: The image may be interfered by various noises during the acquisition process, such as electronic noise, environmental noise, etc. Filtering can effectively remove or reduce these noises and improve image quality. Median filtering is a nonlinear digital filtering technology that is often used to remove noise in images, especially salt and pepper noise, while maintaining the edge details of the image. The basic principle of median filtering is to replace each pixel value in the image with the median of the pixel values ​​in its neighborhood. The calculation formula for median filtering is as follows:

[0052] g(x,y)=med{f(x i ,y j )}

[0053] Among them, g(x,y) is the pixel value after median filtering, f(x i ,y j ) is the pixel value before filtering.

[0054] S2, annotate the cable sample images to identify the defect locations and types.

[0055] S3, using the labeled cable sample images to train the cable surface defect detection model based on the SSD network, where the ResNet50 network is used as the backbone network to replace the VGG16 network in the original SSD network to extract and classify defect features. The principle is as follows Figure 2 shown.

[0056] In this embodiment, conv2_x, conv3_x, and conv4_x in the ResNet50 network are extracted as feature layers and fused, and the last fully connected layer in the original SSD network is removed; the fused feature layer is used as the first convolution layer, and then 6 convolution layers with different convolution kernel sizes are added in sequence behind the convolution layer of the ResNet50 network to generate feature maps of different scales. Feature maps of different scales are used to detect targets of different sizes. The general input image format of SSD is a 300×300 three-color RGB image, and each convolution layer uses a corresponding convolution kernel. The sizes of these 7 convolution kernels are (150, 150), (75, 75), (38, 38), (19, 19), (10, 10), (3, 3), and (1, 1).

[0057] At each convolutional layer, a certain number of anchor boxes are generated based on the size of the feature map and the size of the expected detection target. The number and size of these anchor boxes vary depending on the feature layer to accommodate detection of objects of different sizes. The steps for generating anchor boxes are as follows:

[0058] Step 1: Define the anchor box scale. The anchor box scale calculation formula is as follows:

[0059]

[0060] Among them, S k Indicates the proportion of the anchor box size in the image, S min Indicates the minimum size ratio of the anchor box, S max represents the maximum size ratio of the anchor box, and m represents the number of feature layers.

[0061] Step 2: Define the aspect ratio of the anchor box α r , usually set to α r ∈{1,2,3,1 / 2,1 / 3}, calculate the width and height of the anchor box based on the aspect ratio of the anchor box:

[0062]

[0063] Among them, w k Indicates the width of the anchor box, h k Indicates the height of the anchor box;

[0064] Step 3: Calculate the center coordinates of each anchor box:

[0065]

[0066] Among them, |f k | is the size of the kth feature map, a,b∈{0,1,2,…,|f k -1|}.

[0067] After all feature layers have completed the prediction, the anchor box with the highest prediction score is selected from each feature layer, and the highly overlapping anchor boxes are excluded through the NMS algorithm to reduce redundancy and false detection. After NMS processing, the retained anchor boxes are used as prediction boxes to form the final detection results, which include the location, confidence and category of the target.

[0068] In this way, the improved SSD object detection algorithm can take advantage of the powerful feature extraction capabilities of ResNet50 to improve the detection performance of objects of different sizes, especially in detecting small objects. At the same time, the deep features of ResNet50 also provide rich semantic information for detecting large objects.

[0069] The network is trained using the labeled data set. During the training process, the network parameters are optimized using the loss function and optimizer. The loss function of the cable surface defect detection model consists of positioning loss and confidence loss, and its formula is:

[0070]

[0071] Among them, L conf(x,c) represents the confidence loss, which is used to calculate the difference between the confidence of the predicted category and the true category; L loc (x, l, g) is the positioning loss, which is used to calculate the distance difference between the predicted box and the true box; N represents the number of positive samples, x represents the matching result between the predicted box and the true box, c represents the category confidence prediction value, l represents the position prediction value of the predicted box, g represents the position parameter of the true box, and α is the weight coefficient.

[0072] The operating system of the experimental platform equipped in this invention is Windows 11. In terms of hardware configuration, the CPU is i5-12500H, the GPU is NVIDIA GeForce RTX 3050Ti, 8GB of memory, the software environment is Anaconda, CUDA 11.6, and Python 3.8, PyTorch 1.8.2 version and multiple algorithm target detection frameworks are installed.

[0073] In the process of model training, the optimizer can be used to train the data to learn the ability to distinguish objects of different categories, and evaluate it on the verification data to ensure that the model will not overfit while maintaining good generalization ability. The present invention uses the SGD optimizer to train and optimize the improved SSD algorithm model, including setting the learning rate to 0.001, the momentum to 0.9, the number of iterations to 200, the batch size to 3, and the weight decay rate to 0.0001 for the SGD optimizer to train the model.

[0074] S4, inputting the image to be detected into the trained cable surface defect detection model, detecting and determining the location, size and type of the defect comprehensive model, and realizing automatic detection of cable surface defects.

[0075] This improved SSD target detection method not only improves the efficiency and accuracy of power cable defect detection, but also plays an important role in promoting the intelligentization of power systems.

[0076] Table 1 shows the performance comparison of four different object detection models for cable surface defect detection. These models include Faster R-CNN, YOLO v3, SSD, and improved SSD.

[0077] Table 1 Performance comparison of different algorithms

[0078]

[0079] The table shows the average accuracy of each model in detecting different defects on the cable surface, and the threshold is set to 0.5. mAP represents the average accuracy of the model in all categories, and Recall represents the recall rate of the model, that is, the proportion of positive samples correctly identified by the model to all true positive samples. From the data in the table, it can be seen that the mAP of the improved SSD model of the present invention is the highest, reaching 84.64%, and the mAP of Faster R-CNN is the lowest, which is 79.85%. The AP values ​​of concave and convex surface defects are relatively high in each detection model, and these two surface defects are easy to identify. Compared with the traditional SSD target detection algorithm, the improved SSD target detection algorithm combined with the ResNet50 network as a feature extractor has improved the performance detection of various surface defects.

[0080] The PR curve reflects the relationship between precision P and recall R. The closer the PR curve is to the upper right, the better the model performance. Figure 3 It can be seen that the improved SSD model of the present invention has the best accuracy compared with other models, followed by Faster R-CNN, and SSD is close to YOLO v3.

[0081] The present invention proposes a method, device and medium for detecting surface defects of power cables based on deep learning. The YOLO model and the Faster R-CNN model are both currently commonly used target detection algorithms in deep learning, but the YOLO model is not friendly enough to small target detection and has low accuracy, and the Faster R-CNN model is accurate but has a slow operating speed. Therefore, the present invention adopts the SSD single-stage detection method with good accuracy and speed, and on this basis, uses ResNet50 as the backbone network to replace the VGG16 network in the SSD to improve the algorithm. The improved SSD model has improved accuracy (AP) in the detection of various defects on the cable surface to varying degrees, and the increase in recall rate (Recall) means a reduction in missed detections, that is, ensuring that all positive samples are detected as much as possible.

[0082] Example 2

[0083] The above is an introduction to a method embodiment. The following is a further explanation of the solution of the present invention through an apparatus embodiment.

[0084] This embodiment provides a power cable surface defect detection device based on deep learning, including a memory, a processor, and a program stored in the memory, and when the processor executes the program, the method described in the above embodiment 1 is implemented.

[0085] In a preferred embodiment, Figure 4 As shown, the device comprises:

[0086] 1) Image acquisition module, used to collect images of surface defects of power cables;

[0087] 2) Image processing module, used for preprocessing and image annotation of collected images;

[0088] 3) Algorithm improvement module, used to build a cable surface defect detection model based on SSD network and improve the original SSD network;

[0089] 4) Defect detection module, used to identify and classify defects in the image;

[0090] 5) Defect warning module, used to issue a warning when a defect that may affect the normal operation of the power cable is detected.

[0091] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0092] Example 3

[0093] A schematic diagram of the structure of an electronic device used in the present invention is shown in FIG. Figure 5 As shown, including:

[0094] Memory 501, used to store program code of cable defect detection algorithm, including software for image acquisition, preprocessing, feature extraction, defect identification and classification;

[0095] Processor 502 runs a deep learning model (such as an improved SSD model) to perform feature extraction and defect location.

[0096] Processor 502 may be a central processing unit (CPU), which may perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In RAM, various programs and data required for device operation may also be stored. CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0097] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0098] The processing unit performs the various methods and processes described above, such as methods S1 to S4. For example, in some embodiments, methods S1 to S4 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S4 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S4 in any other appropriate manner (e.g., by means of firmware).

[0099] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0100] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0101] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0102] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for detecting surface defects of power cables based on deep learning, characterized in that: The following steps are involved: Acquire and preprocess images of cable samples with different surface defects; Annotate cable sample images to identify defect locations and types; The labeled cable sample images are used to train the cable surface defect detection model based on the SSD network. The ResNet50 network is used as the backbone network to replace the VGG16 network in the original SSD network to extract and classify defect features. The image to be inspected is input into the trained cable surface defect detection model to automatically detect defects on the cable surface.

2. A method for detecting surface defects of power cables based on deep learning according to claim 1, characterized in that: The use of the ResNet50 network as the backbone network to replace the VGG16 network in the original SSD network is specifically as follows: The conv2_x, conv3_x, and conv4_x in the ResNet50 network are extracted as feature layers and fused, and the last fully connected layer in the original SSD network is removed; the fused feature layer is used as the first convolution layer, and then 6 convolution layers with different convolution kernel sizes are added in sequence behind the convolution layer of the ResNet50 network to generate feature maps of different scales, which are used to detect targets of different sizes.

3. A method for detecting surface defects of power cables based on deep learning according to claim 2, characterized in that: At each convolutional layer, an anchor box is generated based on the size of the feature map and the size of the expected detection object.

4. A method for detecting surface defects of power cables based on deep learning according to claim 3, characterized in that: The steps for generating the anchor point frame are as follows: Define the anchor box scale. The anchor box scale calculation formula is as follows: Among them, S k Indicates the proportion of the anchor box size in the image, S min Indicates the minimum size ratio of the anchor box, S max represents the maximum size ratio of the anchor box, and m represents the number of feature layers; Define the anchor box aspect ratio α r , calculate the width and height of the anchor box based on the aspect ratio of the anchor box: Among them, w k Indicates the width of the anchor box, h k Indicates the height of the anchor box; Calculate the center coordinates of each anchor box: Among them, |f k | is the size of the kth feature map, a,b∈{0,1,2,…,|f k -1|}.

5. The method for detecting surface defects of power cables based on deep learning according to claim 3 is characterized in that: After the cable surface defect detection model completes the prediction in all feature layers, it selects the anchor box with the highest prediction score from each feature layer, and excludes highly overlapping anchor boxes through the NMS algorithm. The retained anchor boxes serve as the prediction boxes to form the final detection results, which include the location, confidence and category of the target.

6. A method for detecting surface defects of power cables based on deep learning according to claim 5, characterized in that: The loss function of the cable surface defect detection model consists of positioning loss and confidence loss, and its formula is: Among them, L conf (x,c) represents the confidence loss, which is used to calculate the difference between the confidence of the predicted category and the true category; L loc (x, l, g) is the positioning loss, which is used to calculate the distance difference between the predicted box and the true box; N represents the number of positive samples, x represents the matching result between the predicted box and the true box, c represents the category confidence prediction value, l represents the position prediction value of the predicted box, g represents the position parameter of the true box, and α is the weight coefficient.

7. The method for detecting surface defects of power cables based on deep learning according to claim 1, characterized in that: The pre-processing comprises the following steps: CLAHE image enhancement processing: the cable sample image is divided into small blocks, and histogram equalization is applied locally to enhance the local contrast by limiting the contrast while avoiding enhancing the background noise; Image filtering: Use median filtering to filter the image that has undergone image enhancement processing.

8. A power cable surface defect detection device based on deep learning, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 7 is implemented.

Images