A target detection method and related device based on circuit breaker test

Abstract

The present application discloses a target detection method based on a circuit breaker test and a related device, the method comprising: constructing a target contact detection model based on a circuit breaker test according to YOLOv4 and Mobiledets, the target contact detection model comprising a Mobiledets layer; training the target contact detection model using a preset test training image set to obtain an optimized contact detection model; detecting the current image to be tested using the optimized contact detection model to obtain a predicted image, the predicted image comprising a target prediction box, confidence and prediction box coordinates. Lightweight Mobiledets is used as the feature extraction network layer of YOLOv4 to directly reduce the number of parameters of the target contact detection model, thereby reducing the calculation pressure and improving the detection speed of the model. Therefore, the present application can solve the technical problem that the existing target detection method cannot avoid a large amount of calculation, resulting in a slow actual detection speed.

Description

Technical Field

[0001] The present application relates to the technical field of circuit breaker testing, and in particular to a target detection method based on circuit breaker testing and a related device. Background Art

[0002] Accurate and reliable wiring is one of the key technologies for the practical application of circuit breaker test robots. It requires the robot's target detection algorithm for identifying and locating circuit breaker contacts to have strong generalization ability and high accuracy, so that it can accurately and reliably grasp the test wire clamp and connect the circuit breaker contacts in different test sites (backgrounds) or under the influence of background disturbances such as people walking.

[0003] At present, the single-stage target detection algorithm [1] used in target detection uses a trained convolutional neural network to recognize images and locate targets with rectangular boxes. The main single-stage target detection algorithms include SSD, RetinaNet, YOLO, etc. However, in addition to the large number of images that need to be processed, the existing convolutional neural network itself contains a large amount of calculation. The resulting calculation pressure leads to a slow actual detection speed and cannot meet the target detection requirements. Summary of the invention

[0004] The present application provides a target detection method based on circuit breaker test and related devices, which are used to solve the technical problem that the existing target detection method cannot avoid a large amount of calculation, resulting in a slow actual detection speed.

[0005] In view of this, the first aspect of the present application provides a target detection method based on a circuit breaker test, comprising:

[0006] A target contact detection model based on a circuit breaker test is constructed according to YOLOv4 and Mobiledets, wherein the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer, and a non-maximum suppression layer;

[0007] The target contact detection model is trained using a preset test training image set to obtain an optimized contact detection model, wherein the preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set;

[0008] The current image to be tested is detected by using the optimized contact detection model to obtain a predicted image, which includes a target prediction box, a confidence level, and prediction box coordinates.

[0009] Preferably, the method of using a preset test training image set to train the target contact detection model to obtain an optimized contact detection model further includes:

[0010] The target contacts in the acquired initial training image set of the circuit breaker test robot wiring are labeled to obtain label information, where the label information includes a labeling frame, frame size, and frame position;

[0011] Obtaining preset background slices above, below, to the left and to the right of the marked frame respectively based on the frame size and the frame position;

[0012] The initial training image set is subjected to an image enhancement operation according to preset background slices in a pasting manner to obtain a preset test training image set.

[0013] Preferably, the performing image enhancement operation on the initial training image set according to the preset background slices in a pasting manner to obtain the preset test training image set includes:

[0014] sorting the training images in the initial training image set;

[0015] The preset background slice corresponding to the current training image is resized according to the sorting order and then pasted on the background area corresponding to the next training image to obtain a preset test training image set.

[0016] Preferably, the method detects the current image to be tested by the optimized contact detection model to obtain a predicted image, wherein the predicted image includes a target prediction box, a confidence level and a prediction box coordinate, and then further includes:

[0017] The action target point coordinates of the wiring robot during the circuit breaker test are calculated according to the target prediction frame and the prediction frame coordinates.

[0018] A second aspect of the present application provides a target detection device based on a circuit breaker test, comprising:

[0019] A model building unit, used for building a target contact detection model based on a circuit breaker test according to YOLOv4 and Mobiledets, wherein the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer and a non-maximum suppression layer;

[0020] A model training unit, used to train the target contact detection model using a preset test training image set to obtain an optimized contact detection model, wherein the preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set;

[0021] The target detection unit is used to detect the current image to be detected by using the optimized contact detection model to obtain a predicted image, wherein the predicted image includes a target prediction box, a confidence level and prediction box coordinates.

[0022] Preferably, it also includes:

[0023] A labeling unit is used to label the target contacts in the acquired initial training image set of the circuit breaker test robot wiring to obtain label information, wherein the label information includes a labeling frame, a frame size, and a frame position;

[0024] A slicing unit, for obtaining preset background slices above, below, left and right of the marked box respectively based on the box size and the box position;

[0025] The pasting unit is used to perform an image enhancement operation on the initial training image set according to the preset background slices in a pasting manner to obtain a preset test training image set.

[0026] Preferably, the pasting unit is specifically used for:

[0027] sorting the training images in the initial training image set;

[0028] The preset background slice corresponding to the current training image is resized according to the sorting order and then pasted on the background area corresponding to the next training image to obtain a preset test training image set.

[0029] Preferably, it also includes:

[0030] A coordinate calculation unit is used to calculate the coordinates of the action target point of the wiring robot during the circuit breaker test according to the target prediction frame and the prediction frame coordinates.

[0031] The present application also provides a target detection device based on a circuit breaker test, the device comprising a processor and a memory;

[0032] The memory is used to store program code and transmit the program code to the processor;

[0033] The processor is used to execute the target detection method based on circuit breaker test described in the first aspect according to the instructions in the program code.

[0034] The present application also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program codes, and the program codes are used to execute the target detection method based on circuit breaker test described in the first aspect.

[0035] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0036] In the present application, a target detection method based on a circuit breaker test is provided, including: constructing a target contact detection model based on the circuit breaker test according to YOLOv4 and Mobiledets, the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer and a non-maximum suppression layer; using a preset test training image set to train the target contact detection model to obtain an optimized contact detection model, the preset test training image set is obtained by background pasting enhancement processing of the initial training image set; detecting the current image to be tested by optimizing the contact detection model to obtain a predicted image, the predicted image including a target prediction box, a confidence and a prediction box coordinate.

[0037] The target detection method based on the circuit breaker test provided by the present application improves the network on the basis of YOLOv4, adopts lightweight Mobiledets as the feature extraction network layer of YOLOv4, directly reduces the number of parameters of the target contact detection model, thereby reducing the calculation pressure and improving the detection speed of the model; in order to ensure the accuracy and stability of the lightweight detection model, the initial training image set is enhanced by background pasting, the scene categories of the training images are increased, and the generalization ability of the detection model is improved. Therefore, the present application can solve the technical problem that the existing target detection method cannot avoid the large amount of calculation, resulting in a slow actual detection speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic flow chart of a target detection method based on a circuit breaker test provided in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of the structure of a target detection device based on a circuit breaker test provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of the CSPDarknet-53 network structure of YOLOv4 provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the Mobiledets network structure provided in an embodiment of the present application;

[0042] Figure 5 A schematic diagram of the training process of the target contact detection model provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of an image enhancement process based on background pasting provided in an embodiment of the present application;

[0044] Figure 7 The loss function of the model training provided in the embodiment of the present application changes with the training algebra;

[0045] Figure 8 The detection accuracy of the model training provided in the embodiment of the present application changes with the number of training generations. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0047] For easier understanding, see Figure 1 , an embodiment of a target detection method based on a circuit breaker test provided by the present application includes:

[0048] Step 101: construct a target contact detection model based on a circuit breaker test according to YOLOv4 and Mobiledets. The target contact detection model consists of a Mobiledets layer, a multi-scale fusion layer, and a non-maximum suppression layer.

[0049] The YOLOv4 target detection algorithm is used to detect the contact target of the circuit breaker. Its model framework is mainly composed of feature network, multi-scale fusion and non-maximum suppression (NMS). The feature network of YOLOv4 is CSPDdarknet-53, which is composed of N convolutional blocks, each of which contains a convolutional layer (Conv), a normalization layer (BN) and a non-linear activation layer (Act). The image texture, color and other features extracted by the feature network are important bases for the network to predict the target; the feature maps of the three scales output by the feature network are multi-scale fused to predict targets of different sizes; the non-maximum suppression (NMS) algorithm is used to eliminate redundant target detection frames and obtain the best target detection frame.

[0050] However, the CSPDarknet-53 in the YOLOv4 target detection algorithm consists of 5 cross-stage partial network CSP (CrossStage Partial, CSP) modules, with a total of 76 standard convolution blocks. Each time the feature map passes through a CSP module, its dimension will be halved, and the number of channels will be doubled. This results in a large amount of overall algorithm calculation, prolonged reasoning time, and reduced detection efficiency. Therefore, this embodiment uses Mobiledets instead of the feature network layer in YOLOv4 for feature extraction.

[0051] Mobiledets is an object detection model used in mobile and embedded devices. To meet the needs of edge computing, its feature network adopts a lightweight design and has the advantage of extremely short inference time. In order to determine the advantages of this improved algorithm, a comparison of computational effort can be performed. First, the CSPDdarknet-53 network is shown in Figure 1. Figure 3 , if the addition operation is ignored, the number of feature extraction operations of any convolution block can be expressed as:

[0052] K ci =D ci ×C ci ×M ci ×D pi

[0053] Among them, D ci =d ci ×d ci , where d ci =3 is the convolution kernel dimension of any convolution block i, i = 1, 2, ..., n; C ci is the number of convolution kernels or channels; M ci is the number of channels of the input feature map; D pi The number of horizontal and vertical slides for each convolution kernel. The total number of feature extraction calculations of the feature network can be expressed as:

[0054]

[0055] Where n = 76 is the number of convolutional blocks contained in the feature network; K pi is the total number of parameters of the convolution output feature map; BN(·) means the normalization calculation of the convolution output feature map, Act(·) is the nonlinear activation calculation of the normalized feature map, and the number of calculations is linearly related to the dimension of the convolution output feature map.

[0056] Therefore, the inference time of the total number of calculations for feature extraction of the feature network CSPDarknet-53 is:

[0057] T c =K c ×T0

[0058] In the formula, T0 is the single feature extraction calculation time under different floating point standards. From the above formula, we can calculate that the average inference time of each image of the feature network CSPDarknet-53 under the floating point standard FP32 is about 17.89ms.

[0059] Next is the improved target contact detection model, see Figure 4, that is, the feature network replaces the module Mobiledets. At the same time, it is also necessary to adjust the feature map branches and channels of the network output to make it compatible with the subsequent network of YOLOv4, that is, the multi-scale fusion layer and the non-maximum suppression layer. The adjusted Mobiledets mainly consists of 4 IBN modules, 17 Fused modules and 3 Tucker modules. The network structure is as follows Figure 4 , where the IBN module contains one depth-wise separable convolution block and one standard convolution block (1×1 convolution kernel). The number of feature extraction calculations for this module is:

[0060] K dj =D dj ×M dj ×D pj +M cj ×C cj ×D pj

[0061] Where D dj =d dj ×d dj , where d dj =3 is the dimension of the convolution kernel of the depthwise separable convolution block of any IBN module j; j = 1, 2, ..., m; C cj is the number of convolution kernels or channels of the standard convolution block; M dj 、M cj are the number of feature map channels of the input depth-separable convolutional block and the standard convolutional block respectively; D pj Number of horizontal and vertical slides for each convolution kernel.

[0062] The Fused module and Tucker module contain two 1×1 standard convolution blocks with different parameters, and the number of feature extraction calculations are:

[0063]

[0064] Where D sk , D eq They are 1×1 convolution blocks, and the value is 1; s and e are the feature map channel scaling factors, s is 4 or 8, and e is 0.75 or 0.25.

[0065] If the input feature map dimension and number of channels remain unchanged, that is, D pk , D pq and M sk 、M eq The total number of calculations for extracting features from the feature network Mobiledets is fixed as follows:

[0066]

[0067] Where n s=20, n e =3 are the number of 1×1 standard convolution blocks with two different parameters; n h =24 is the number of 3×3 convolution blocks; n l is the number of depth-wise separable convolutional blocks; n r =n s +n e +n h +n l is the total number of feature network convolution blocks. In the above formula, the number of feature extraction calculations of the depth-separable convolution block K dj About K ci 1 / 9 of the feature extraction calculation times K of the 1×1 standard convolution block with two different parameters sk , K eq K ci The total number of calculations for extracting features by Mobiledets is much smaller than that of CSPDarknet-53. The above formula shows that the reasoning time of the feature network Mobiledets for each image is about 7.1ms, which is only 40% of the reasoning time of CSPDarknet-53.

[0068] By comparison, it can be found that the improved target contact detection model has smaller parameters and shorter calculation time, that is, shorter reasoning time, which can speed up the response speed of the model and improve detection efficiency.

[0069] Step 102: Use a preset test training image set to train the target contact detection model to obtain an optimized contact detection model. The preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set.

[0070] Before using the model for detection, it is necessary to use a training image set for training and optimize the model performance. Only after reaching a certain standard can the optimized contact detection model be obtained. The preset test training image can be a historical data acquired in advance, that is, an initial training image. At the same time, the historical data can be subjected to basic redundancy removal and noise removal preprocessing operations as needed; then image enhancement is performed and the data set is expanded, so that the generalization performance of the model training can be guaranteed. The enhancement method selected in this embodiment is background pasting, that is, increasing the diversity of the detection environment, so that the detection model can adapt to target detection in more environments and have stronger anti-interference ability.

[0071] For the specific model training process, please refer to Figure 5After data enhancement of the image data of the training data set, the target contact detection model is pre-trained to obtain the pre-training results; the loss is calculated based on the pre-training results and the label information of the training data set, and then the weight of the model is adjusted and updated according to the loss results, and then it is determined whether the model training meets the standards. If so, an optimized contact detection model can be obtained.

[0072] Furthermore, step 102, before that, also includes:

[0073] The target contacts in the acquired initial training image set of the circuit breaker test robot wiring are labeled to obtain label information, where the label information includes a labeling frame, frame size, and frame position;

[0074] Obtaining preset background slices above, below, to the left and to the right of the marked frame respectively based on the frame size and the frame position;

[0075] The image enhancement operation is performed on the initial training image set according to the preset background slices in a pasting manner to obtain a preset test training image set.

[0076] This step is the specific acquisition process of the preset test training image set. The background of the original training image is continuously generated and sliced, and pasted to the background area of ​​other images to form a new background, so that the expanded image training samples have sufficient background features.

[0077] In this embodiment, the detection target is the circuit breaker contact, which is marked with a label box, and the size and position information of the label box are given; in addition, during the model training process, the label information also includes confidence. By training the target detection model based on the labeled training image set, the training of the model can be controlled and adjusted, thereby improving the detection performance of the model.

[0078] See also Figure 6 , the top, bottom, left and right sides of the marked box object can each obtain a preset background slice, which can be represented as p t 、p b 、p l 、p r ; The label box uses its upper left corner coordinates (x1, y1) and lower right corner coordinates (x2, y2) to describe its position.

[0079] Four preset background slices t 、p b 、p l 、p r The size and position of are random, but cannot contain any foreground features, that is, the area within the annotation box cannot be included in the preset background slice; the quantitative description is p tThe bottom boundary is higher than y1, p b The upper boundary is lower than y2, p l The right boundary is less than x1, p r The left boundary of is greater than x2. Randomly slice the area as described and extract the four slices obtained, as shown in Figure 6 Shown on the left.

[0080] Further, the initial training image set is subjected to image enhancement operation according to the preset background slices in a pasting manner to obtain a preset test training image set, including:

[0081] Sort the training images in the initial training image set;

[0082] According to the sorting order, the preset background slice corresponding to the current training image is resized and pasted on the background area corresponding to the next training image to obtain a preset test training image set.

[0083] See also Figure 6 , the preset background slice extracted from one training image can be resized according to the vacant size after the slice is extracted from another training image and then pasted into the corresponding background area to form a new background image, thereby realizing the enhancement operation of the training image. If the two training images here are in sequence, the preset background slice of the current training image is pasted to the next training image, and the background pasting of all training images is completed in this way. The preset background slice of the last training image can be pasted to the first training image to realize the closed-loop operation.

[0084] This embodiment only provides an example scheme for background pasting. Other background pasting schemes may also be designed according to actual conditions, such as jump pasting, that is, the preset background slice of the current training image is resized and pasted on the background area of ​​the second, third or nth training image to achieve image enhancement. The specific operation process is similar to the above scheme and will not be repeated here.

[0085] Step 103: Detect the current image to be detected by optimizing the contact detection model to obtain a predicted image, which includes a target prediction box, a confidence level, and prediction box coordinates.

[0086] The target prediction box is the area where the circuit breaker contacts are selected, and the prediction box coordinates can reflect the position information of the target after it is selected by the marked box. Accurate target prediction can provide reliable support for subsequent robot positioning and improve the robot positioning accuracy of circuit breaker tests.

[0087] Furthermore, step 103 further includes:

[0088] The target point coordinates of the wiring robot during the circuit breaker test are calculated based on the target prediction frame and the prediction frame coordinates.

[0089] The training images are all acquired by cameras. Based on the focal length and optical principles of the camera itself and the coordinates of the target prediction frame in the acquired image, the specific coordinates of the robot's action point can be calculated, thereby achieving accurate positioning and ensuring the reliability of the robot's wiring.

[0090] In order to verify the effect of the solution provided by this embodiment, a simulation test of the accuracy of the circuit breaker contact target detection model is provided; the generalization ability, accuracy and stability of the model are verified through simulation tests. The circuit breaker contact target detection model is trained using image background expansion training samples (113,400 images). After each generation of training, the trained target detection model is used to predict the image test samples, and the detection accuracy of the circuit breaker contacts, that is, the accuracy of the prediction box position, is calculated using the prediction box and true box information. The loss function and the detection accuracy obtained by training according to the above process are as follows: Figure 7 , Figure 8 shown.

[0091] Depend on Figure 7 It can be seen that in the early stage of training, the model is poorly adapted to the differences in the background of the training image dataset, and the loss function value is large; as the number of training generations increases, the decrease in the loss function value gradually slows down; after training for 300 generations, the difference in the background of the training image has little effect on the loss function value and converges.

[0092] Depend on Figure 8 It can be seen that the value of the regression loss function ciou-loss of the prediction box gradually decreases with the increase of training generations, while the detection accuracy gradually increases; after training to 300 generations, the accuracy d-accuracy and c-accuracy of the prediction box position and size output by the model both reach 99.9%.

[0093] The simulation test results also show that the circuit breaker contact target detection model trained with image samples expanded by the background data enhancement method is not affected by the test sample background and has strong generalization ability and high accuracy.

[0094] In addition, test cases based on the improved model, namely the wiring of each test item and the total wiring time of the target contact detection model are also given. From Table 1, it can be seen that the wiring time of the algorithm model of this embodiment is less than the wiring time based on the standard YOLOv4 target detection model.

[0095] Table 1 Test results of wiring time of circuit breaker test robot for different test items

[0096]

[0097] The test results show that the target detection method for circuit breaker test based on background data enhancement and improved YOLOv4 proposed in this application not only has high wiring accuracy but also short wiring time, which can meet the technical requirements of the application of circuit breaker robot test platform.

[0098] The target detection method based on the circuit breaker test provided in the embodiment of the present application improves the network on the basis of YOLOv4, adopts lightweight Mobiledets as the feature extraction network layer of YOLOv4, directly reduces the number of parameters of the target contact detection model, thereby reducing the calculation pressure and improving the detection speed of the model; in order to ensure the accuracy and stability of the lightweight detection model, the initial training image set is enhanced by background pasting, the scene categories of the training images are increased, and the generalization ability of the detection model is improved. Therefore, the embodiment of the present application can solve the technical problem that the existing target detection method cannot avoid a large amount of calculation, resulting in a slow actual detection speed.

[0099] For easier understanding, see Figure 2 The present application provides an embodiment of a target detection device based on a circuit breaker test, comprising:

[0100] A model building unit 201 is used to build a target contact detection model based on a circuit breaker test according to YOLOv4 and Mobiledets, wherein the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer, and a non-maximum suppression layer;

[0101] The model training unit 202 is used to train the target contact detection model using a preset test training image set to obtain an optimized contact detection model, wherein the preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set;

[0102] The target detection unit 203 is used to detect the current image to be detected by optimizing the contact detection model to obtain a predicted image, which includes a target prediction box, a confidence level and prediction box coordinates.

[0103] Furthermore, it also includes:

[0104] A labeling unit 204 is used to label the target contacts in the acquired initial training image set of the circuit breaker test robot wiring to obtain label information, where the label information includes a labeling frame, frame size, and frame position;

[0105] A slicing unit 205, configured to obtain preset background slices above, below, left and right of the marked frame respectively based on the frame size and frame position;

[0106] The pasting unit 206 is used to perform an image enhancement operation on the initial training image set according to the preset background slices in a pasting manner to obtain a preset test training image set.

[0107] Furthermore, the pasting unit 206 is specifically used for:

[0108] Sort the training images in the initial training image set;

[0109] According to the sorting order, the preset background slice corresponding to the current training image is resized and pasted on the background area corresponding to the next training image to obtain a preset test training image set.

[0110] Furthermore, it also includes:

[0111] The coordinate calculation unit 207 is used to calculate the coordinates of the target point of the wiring robot during the circuit breaker test according to the target prediction frame and the prediction frame coordinates.

[0112] The present application also provides a target detection device based on a circuit breaker test, the device comprising a processor and a memory;

[0113] The memory is used to store the program code and transmit the program code to the processor;

[0114] The processor is used to execute the target detection method based on the circuit breaker test in the above method embodiment according to the instructions in the program code.

[0115] The present application also provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the target detection method based on circuit breaker test in the above method embodiment.

[0116] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0117] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0118] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0119] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for executing all or part of the steps of the method described in each embodiment of the present application through a computer device (which can be a personal computer, a server, or a network device, etc.). The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0120] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A target detection method based on circuit breaker test, characterized in that: include: A target contact detection model based on a circuit breaker test is constructed according to YOLOv4 and Mobiledets, wherein the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer, and a non-maximum suppression layer; The target contacts in the acquired initial training image set of the circuit breaker test robot wiring are labeled to obtain label information, where the label information includes a labeling frame, frame size, and frame position; Obtaining preset background slices above, below, to the left and to the right of the marked frame respectively based on the frame size and the frame position; The initial training image set is subjected to image enhancement operation according to the preset background slices in a pasting manner to obtain a preset test training image set. The specific enhancement process is as follows: sorting the training images in the initial training image set; According to the sorting order, the preset background slice corresponding to the current training image is resized and pasted to the background area corresponding to the next training image to obtain a preset test training image set; The target contact detection model is trained using a preset test training image set to obtain an optimized contact detection model, wherein the preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set; The current image to be tested is detected by using the optimized contact detection model to obtain a predicted image, which includes a target prediction box, a confidence level, and prediction box coordinates.

2. The target detection method based on circuit breaker test according to claim 1, characterized in that: The optimized contact detection model is used to detect the current image to be tested to obtain a predicted image, wherein the predicted image includes a target prediction frame, a confidence level, and a prediction frame coordinate, and then further includes: The action target point coordinates of the wiring robot during the circuit breaker test are calculated according to the target prediction frame and the prediction frame coordinates.

3. A target detection device based on circuit breaker test, characterized in that: include: A model building unit, used for building a target contact detection model based on a circuit breaker test according to YOLOv4 and Mobiledets, wherein the target contact detection model is composed of a Mobiledets layer, a multi-scale fusion layer and a non-maximum suppression layer; A labeling unit is used to label the target contacts in the acquired initial training image set of the circuit breaker test robot wiring to obtain label information, wherein the label information includes a labeling frame, a frame size, and a frame position; A slicing unit, for obtaining preset background slices above, below, left and right of the marked box respectively based on the box size and the box position; A pasting unit is used to perform an image enhancement operation on the initial training image set according to the preset background slices in a pasting manner to obtain a preset test training image set, and the pasting unit is specifically used to: sorting the training images in the initial training image set; According to the sorting order, the preset background slice corresponding to the current training image is resized and pasted to the background area corresponding to the next training image to obtain a preset test training image set; A model training unit, used to train the target contact detection model using a preset test training image set to obtain an optimized contact detection model, wherein the preset test training image set is obtained by performing background pasting enhancement processing on the initial training image set; The target detection unit is used to detect the current image to be detected by using the optimized contact detection model to obtain a predicted image, wherein the predicted image includes a target prediction box, a confidence level and prediction box coordinates.

4. The target detection device based on circuit breaker test according to claim 3, characterized in that: Also includes: A coordinate calculation unit is used to calculate the coordinates of the action target point of the wiring robot during the circuit breaker test according to the target prediction frame and the prediction frame coordinates.

5. A target detection device based on circuit breaker test, characterized in that: The device comprises a processor and a memory; The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the target detection method based on circuit breaker test according to any one of claims 1-2 according to the instructions in the program code.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the target detection method based on circuit breaker test according to any one of claims 1-2.

Images