CN113888514A - Method and device for detecting defects of ground wire, edge computing equipment and storage medium - Google Patents
Method and device for detecting defects of ground wire, edge computing equipment and storage medium Download PDFInfo
- Publication number
- CN113888514A CN113888514A CN202111177578.0A CN202111177578A CN113888514A CN 113888514 A CN113888514 A CN 113888514A CN 202111177578 A CN202111177578 A CN 202111177578A CN 113888514 A CN113888514 A CN 113888514A
- Authority
- CN
- China
- Prior art keywords
- model
- ground wire
- image data
- preset
- yolov4
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The application relates to a method and a device for detecting defects of a ground wire, computer equipment and a storage medium. The method comprises the steps of obtaining original image data of a power tower area of a power grid; determining position information of a ground wire in original image data of an electric tower area through a preset first identification model; inputting the position information of the ground wire into a preset second identification model, and determining the defect information of the ground wire in the original image data of the electric tower area; and sending a risk warning according to the fault information of the ground wire. According to the scheme, the defect detection of the power grid conducting wire is achieved based on the edge detection, after the original image data of the power tower area are obtained, double recognition is directly carried out through edge computing equipment, the position of the ground wire is determined firstly, then the defect of the ground wire is determined, the lightweight recognition model based on the YOLOV4-tiny network structure can effectively run on the local edge computing equipment, the time consumption of recognition reasoning is reduced, therefore, high-definition image transmission is not needed, and the recognition analysis efficiency of the defect of the ground wire is improved.
Description
Technical Field
The application relates to the field of power distribution of a power grid, in particular to a method and a device for detecting defects of a ground wire, computer equipment and a storage medium.
Background
Due to the constant development of modern industry, the demand for electric power is constantly increasing: in order to stably supply a large amount of power, a high transmission voltage is required. And the wire in the transmission line bears the long-term vibration action such as breeze vibration, fretting wear easily occurs at the wire clamp outlet, and the wire clamp outlet is also a stress concentration area, so that the generation and the expansion of radial cracks of the wire at the wire are caused by the superposition of the wear and the stress concentration effect, the cracks are gradually developed under the action of periodic stress, and finally the fatigue fracture of the wire is caused. The broken strand of the ground wire of the power transmission line is a common defect in operation and maintenance, and the broken strand can affect the current-carrying capacity of the line, cause corona and reduce the mechanical performance of the line. The broken wire of the earth wire is a serious fault which still occurs at present, the difficulty of emergency repair is large, and the power failure time is long.
At present, the defect treatment of the ground wire can be realized through the unmanned aerial vehicle technology. The unmanned aerial vehicle image acquisition is remote control or code intelligent cruise to carry out field data acquisition on key positions of the iron tower, and then manual data analysis is carried out on a server.
Disclosure of Invention
In view of the foregoing, there is a need to provide a method, an apparatus, a computer device and a storage medium for detecting defects of a ground wire, which can effectively monitor defects of a power grid conductive wire in real time.
A method for detecting defects of a ground wire, comprising the following steps:
acquiring original image data of a power tower area of a power grid;
determining position information of a conducting wire and a ground wire in the original image data of the electric tower area through a preset first identification model, wherein the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, and determining the defect information of the ground wires in the original image data of the electric tower area, wherein the preset second recognition model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and sending a risk warning according to the defect information of the ground wire.
In one embodiment, the determining, by presetting the first recognition model, the ground wire position information in the raw image data of the electric tower region includes:
extracting a first feature map and a second feature map corresponding to the original image data of the electric tower region through the preset first identification model;
performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map;
processing the fusion feature map based on an attention mechanism to obtain a classification feature map;
obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm;
and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
In one embodiment, before determining, by presetting the first recognition model, the position information of the ground wire in the raw image data of the electric tower region, the method further includes:
acquiring historical image data containing ground wire equipment and first marking information corresponding to the historical image data, wherein the first marking information is used for marking the body position of the ground wire and the abnormal position of the ground wire in the historical image data;
labeling the historical image data through the first labeling information to obtain first model training data;
and training the initial Yolov4-tiny model based on the first model training data to obtain a preset first recognition model.
In one embodiment, the training of the initial YOLOV4-tiny model based on the first model training data, and the obtaining of the preset first recognition model includes:
dividing the first model training data into training set data, test set data and verification set data;
performing iterative training on the initial Yolov4-tiny model through the training set data, the test set data and the verification set data, and performing model compression processing on the Yolov4-tiny model which is finished each time of iteration in the iterative training process;
when the loss of the training set data and the loss of the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold value, stopping iteration, and taking a YOLOV4-tiny model after current iteration training as a first available model;
and performing marginalization processing on the first available model to obtain a preset first recognition model.
In one embodiment, the marginalizing the first available model, and obtaining the preset first recognition model includes:
converting the first available model to an onnx model;
analyzing the onnx model based on TensorRT to obtain an Engine model;
and carrying out quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
In one embodiment, after the labeling the historical image data by the first labeling information and obtaining first model training data, the method further includes:
extracting a ground wire area image in the first model training based on the position of the ground wire body;
determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first marking information;
marking the ground wire area image according to the abnormal position of the ground wire in the ground wire area image to obtain second model training data;
and training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model.
A device for detecting defects of a ground wire is applied to an edge computing device and comprises:
the data acquisition module is used for acquiring original image data of an electric tower area of a power grid;
the first identification module is used for determining position information of a conducting wire in the original image data of the electric tower area through a preset first identification model, and the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
the second identification module is used for inputting the position information of the ground wires into a preset second identification model and determining the defect information of the ground wires in the original image data of the electric tower area, and the preset second identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and the risk warning module is used for sending risk warning according to the defect information of the ground wire.
In one embodiment, the first identification module is specifically configured to: extracting a first feature map and a second feature map corresponding to the original image data of the electric tower region through the preset first identification model; performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map; processing the fusion feature map based on an attention mechanism to obtain a classification feature map; obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm; and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
An edge computing device comprising a memory and a processor, the memory storing a computer program that when executed by the processor performs the steps of:
acquiring original image data of a power tower area of a power grid;
determining position information of a conducting wire and a ground wire in the original image data of the electric tower area through a preset first identification model, wherein the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, and determining the defect information of the ground wires in the original image data of the electric tower area, wherein the preset second recognition model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and sending a risk warning according to the defect information of the ground wire.
A computer-readable storage medium, on which a computer program is stored which, when executed by a processor, carries out the steps of:
acquiring original image data of a power tower area of a power grid;
determining position information of a conducting wire and a ground wire in the original image data of the electric tower area through a preset first identification model, wherein the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, and determining the defect information of the ground wires in the original image data of the electric tower area, wherein the preset second recognition model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and sending a risk warning according to the defect information of the ground wire.
According to the method and the device for detecting the defects of the ground wire, the edge computing equipment and the storage medium, the original image data of the power tower area of the power grid are obtained; determining position information of a ground wire in original image data of an electric tower area through a preset first identification model; inputting the position information of the ground wire into a preset second identification model, and determining the defect information of the ground wire in the original image data of the electric tower area; and sending a risk warning according to the fault information of the ground wire. According to the scheme, the defect detection of the power grid conducting wire is achieved based on the edge detection, after the original image data of the power tower area are obtained, double recognition is directly carried out through edge computing equipment, the position of the ground wire is determined firstly, then the defect of the ground wire is determined, the lightweight recognition model based on the YOLOV4-tiny network structure can effectively run on the local edge computing equipment, the time consumption of recognition reasoning is reduced, therefore, high-definition image transmission is not needed, and the recognition analysis efficiency of the defect of the ground wire is improved.
Drawings
FIG. 1 is a schematic flow chart illustrating a method for detecting defects of a ground wire in one embodiment;
FIG. 2 is a sub-flow diagram of step 104 of FIG. 1 in one embodiment;
FIG. 3 is a schematic flow chart illustrating steps of constructing a default first identification model according to one embodiment;
FIG. 4 is a sub-flow diagram of step 306 of FIG. 3 in one embodiment;
FIG. 5 is a sub-flow diagram of step 407 in FIG. 4 in one embodiment;
FIG. 6 is a flowchart illustrating steps of constructing a predetermined second recognition model according to an embodiment;
FIG. 7 is a block diagram of an embodiment of a device for detecting defects of a ground wire;
FIG. 8 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In one embodiment, as shown in fig. 1, a method for detecting a ground wire defect is provided, and the method is particularly applied to an edge monitoring device. In this embodiment, the method includes the steps of:
and 102, acquiring original image data of a power tower area of a power grid.
The edge computing means that an open platform integrating network, computing, storage and application core capabilities is adopted on one side close to an object or a data source to provide nearest-end service nearby. The application program is initiated at the edge side, so that a faster network service response is generated, and the basic requirements of the industry in the aspects of real-time business, application intelligence, safety, privacy protection and the like are met. The edge computation is between the physical entity and the industrial connection, or on top of the physical entity. In the embodiment of the present application, the detection of the ground wire defect is directly performed at the position where the original image data of the electric tower area is collected.
Specifically, when conducting ground wire defect detection, it is necessary to acquire original image data of the electric tower area in real time and perform identification and judgment based on the acquired data. In one embodiment, an image capture device on the drone may be employed to obtain raw image data of the tower area of the power grid. At this moment, can install edge computing equipment additional on unmanned aerial vehicle to be connected edge computing equipment and image acquisition equipment, thereby can directly obtain the regional primitive image data of electric tower of electric wire netting in real time, and detect.
104, determining position information of a ground wire in original image data of the electric tower area through a preset first identification model, wherein the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and 106, inputting the position information of the ground wires into a preset second recognition model, determining the defect information of the ground wires in the original image data of the electric tower area, and constructing the preset second recognition model based on a YOLOV4-tiny network structure through a YOLOV4 algorithm.
And step 108, sending a risk warning according to the grounding wire defect information.
Among them, YOLOV4 is an end-to-end target object detection algorithm based on convolutional neural network, which converts the target detection problem into a regression problem, and this method significantly improves the object detection speed. Specifically, the input image is divided into an S × S grid: the mesh is responsible for detecting objects if the true center of the object is within its boundaries. The object is then predicted by a bounding box on each mesh, and the final coordinates of the bounding box and the class probabilities are generated by a regression algorithm. And YOLOV4-tiny is a compressed version of YOLOV 4. Based on YOLOV4, the network structure is simpler, and parameters are reduced, so that the network structure becomes a feasible proposal for development of mobile and embedded devices. YOLOV4-tiny can be used for faster training and faster detection. It has only two YOLO headers and only three of YOLOV4 and has been trained from 29 pre-trained convolutional layers, while YOLOV4 has been trained from 137 pre-trained convolutional layers. The defects of the ground wire particularly mean the defects of broken strands or scattered strands and the like of the ground wire.
In particular, when conducting wire and ground wire defect detection is carried out, detection on edge computing equipment can be achieved based on a neural network model of a light-weight Yolov4-tiny network structure. During detection, detection can also be performed step by step through a cascade algorithm, the algorithm firstly calls a preset first identification model to detect the original image data of the electric tower area, judges the position information of the ground wires in the image, inputs the position information of the ground wires into a preset second identification model, and determines the defect information of the ground wires in the original image data of the electric tower area so as to determine whether the category information of broken strands and scattered strands exist on the ground wires and the position information of the defect of the ground wires if the defect occurs. And then sending a risk warning according to the fault information of the ground wire.
According to the method for detecting the defects of the ground wire, original image data of the power tower area of the power grid are obtained; determining position information of a ground wire in original image data of an electric tower area through a preset first identification model; inputting the position information of the ground wire into a preset second identification model, and determining the defect information of the ground wire in the original image data of the electric tower area; and sending a risk warning according to the fault information of the ground wire. According to the scheme, the defect detection of the power grid conducting wire is achieved based on the edge detection, after the original image data of the power tower area are obtained, double recognition is directly carried out through edge computing equipment, the position of the ground wire is determined firstly, then the defect of the ground wire is determined, the lightweight recognition model based on the YOLOV4-tiny network structure can effectively run on the local edge computing equipment, the time consumption of recognition reasoning is reduced, therefore, high-definition image transmission is not needed, and the recognition analysis efficiency of the defect of the ground wire is improved.
In one embodiment, as shown in FIG. 2, step 104 comprises:
step 201, extracting a first feature map and a second feature map corresponding to the original image data of the electric tower area through a preset first recognition model.
And 203, performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map.
And step 205, processing the fused feature map based on an attention mechanism to obtain a classification feature map.
And step 207, acquiring a prediction box of the classification characteristic map through a GIoU algorithm and an NMS algorithm.
And step 209, determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
Among other things, the main purpose of the upsampling process is to enlarge the original image so that it can be displayed on a higher resolution display device. When the image features are extracted through the preset first recognition model, two features with different sizes are obtained, and in order to better learn the feature information in the image, the smaller feature map can be subjected to upsampling processing, so that the smaller feature map and the larger feature map are fused. The attention mechanism is a resource allocation mechanism, and can be understood that for the original evenly allocated resources, the important units are divided into more units and the unimportant or bad units are divided into less units according to the importance degree of the attention object.
Specifically, in order to enable a detection algorithm to run stably, quickly and in real time on edge equipment, a lightweight network structure needs to be used, and time consumed by model reasoning is reduced, so that a YOLOV4-tiny network structure is adopted in the method, a CSPDarknet53-tiny, a Mish activation function, a Dropblock and other modes are adopted on a YOLOV4s background, a part of adjacent whole area is dropped through dropout, and the network can pay attention to learning of other part features to realize correct classification, so that better generalization is represented. It is a combination of Darknet-19 and Res-net modules. CSPDarknet53 is the CSP added to each large residual block of Darknet 53. The backbone network also uses SPP modules (spatial pyramid pooling) in order to increase the perceived view of the network. The normalized data are transmitted into a neural network for feature extraction, the input of a Yolov4-tiny network is 416 x 3, two feature maps with different sizes are finally obtained through a convolution layer, a BN layer and an activation layer, namely a first feature map of 13 x 13 and a second feature map of 26 x 26, in order to better learn feature information in an image, upsampling processing can be used for the first feature map of 13 x 13, the processed feature maps and the feature map of 26 x 26 are fused to obtain a fused feature map, and the identification accuracy is improved through the method. The YOLOV4-tiny network adopts PAN structure, and the deep characteristic diagram carries stronger semantic characteristic and weaker positioning information. And shallow feature maps carry stronger location information and weaker semantic features. FPN is to transfer deep semantic features to shallow layers, thereby enhancing semantic expression on multiple scales. And PAN conversely conducts the positioning information of a shallow layer to a deep layer, so that the positioning capacity on multiple scales is enhanced. Meanwhile, an attention mechanism is introduced into the network, a channel-based global maximum pooling (global max pooling) and a global average pooling (global average pooling) are firstly performed, and then the 2 results are merged based on channels. Then, after a convolution operation, the dimensionality reduction is 1 channel. And generating a spatial attribute feature by the sigmoid. And finally, multiplying the feature by the input feature of the module to obtain the finally generated classification feature map. And finally, calculating the category by using the extracted classification characteristic diagram, and detecting the small target and the large target respectively by using the detection result of the YOLOV4 through 2 different scales. By GIoU algorithm: the area of the minimum closure region of the two frames (i.e. the area of the minimum frame including both the prediction frame and the real frame) is calculated, IoU is calculated, the proportion of the region of the closure region not belonging to the two frames to the closure region is calculated, and finally IoU is used to subtract the proportion to obtain the GIoU. And finally, obtaining a final prediction frame through an NMS algorithm, wherein the position of the prediction frame in the original image of the electric tower region is the position of the ground wire in the original image data of the electric tower region. Specifically, the identification process of the preset second identification model is similar to that of the preset first identification model, but the final identification result is the position of the fault of the ground wire in the image data corresponding to the position part of the ground wire, that is, the position of the broken strand or the scattered strand of the ground wire. In this embodiment, the identification of the position of the ground wire is performed by presetting the first identification model, so that the stable, fast and real-time operation of the model on the edge device can be effectively ensured, and the detection efficiency of the defects of the ground wire is ensured.
In one embodiment, as shown in fig. 3, before step 104, the method further includes:
step 302, obtaining historical image data containing the ground wire device and first labeling information corresponding to the historical image data, wherein the first labeling information is used for labeling the body position of the ground wire and the abnormal position of the ground wire in the historical image data.
And step 304, labeling the historical image data through the first labeling information to obtain first model training data.
Step 306, training the initial YOLOV4-tiny model based on the first model training data to obtain a preset first recognition model.
Specifically, before the identification of the position information of the ground wire is performed by presetting the first identification model, training of the preset first identification model needs to be completed, and at this time, the historical image data including the ground wire equipment and the first annotation information corresponding to the historical image data can be acquired and annotated, so that the historical image data with annotations is obtained. And then taking the marked historical image data as model training data, and carrying out supervised training on the initial Yolov4-tiny model to obtain a final preset first recognition model. In a specific embodiment, an image containing ground wire equipment can be screened out from historical image data acquired by the unmanned aerial vehicle, and an image with low resolution and serious distortion is removed. And then, marking the ground wire body in the image and broken strands and loose strands on the ground wires by using a marking tool, storing the ground wire body, broken strands and loose strands in a VOC data format into an xml file, wherein the xml file contains category information and position information, and converting the VOC data format into a YOLOV4 data format to serve as first model training data. And divided into 80% training set, 10% testing set, 10% validation set. The training of the model can be completed through the data. In the embodiment, the preset first recognition model can be effectively obtained by constructing the training data, and the recognition accuracy of the model is ensured.
In one embodiment, as shown in FIG. 4, step 306 comprises:
step 401, the first model training data is divided into training set data, test set data and verification set data.
And 403, performing iterative training on the initial Yolov4-tiny model through training set data, test set data and verification set data, and performing model compression processing on the Yolov4-tiny model finished in each iteration in the iterative training process.
And 405, when the losses of the training set data and the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold, stopping iteration, and taking the Yolov4-tiny model trained by the current iteration as a first available model.
Step 407, performing marginalization processing on the first available model to obtain a preset first recognition model.
Specifically, after the first model training data is obtained, the model training data can be split into a training set, a testing set and a verification set, wherein the training set is used for training an initial model, the verification set is used for searching for the optimal parameters of the model, the testing set is used for searching for the generalization errors of the trained model, the initial YOLOV4-tiny model is iteratively trained through the training set data, the testing set data and the verification set data, and the iterative training process can specifically refer to the position information of the ground wire in the original image data of the secondary tower region of the preset first recognition model. In each iterative training, pruning and compressing quantization operations can be carried out on the model obtained by the iterative training, the size of the model is reduced, the model can run more efficiently, and then the model is tested, so that the size of the model is compressed under the condition that the precision is hardly reduced. Model pruning is the pruning of neurons that have little importance to the prediction, but have little effect. The model compression quantization is a process of approximating the floating point model weight of continuous values (or a large number of possible discrete values) or tensor data flowing through the model (generally Int 8) to a plurality of (or less) discrete values in a fixed-point mode with low inference precision loss, and the size of the FP16 quantization model can be reduced to half of the original size, so that the memory occupation and the power consumption are reduced, and the inference speed is increased. In the continuous iterative training process, when the losses of the training set and the verification set are reduced until the training set and the verification set are stable, and the accuracy of the verification set is improved to a preset accuracy threshold, the set iteration target is considered to be reached, the iteration is stopped, and the finally obtained Yolov4-tiny model is used as a first available model. And performing marginalization processing on the first available model to obtain a preset first recognition model, so that the first recognition model is more suitable for being used on edge computing equipment. In this embodiment, the recognition accuracy of the preset first recognition model can be effectively ensured through processing such as iterative training.
In one embodiment, as shown in FIG. 5, step 407 comprises:
step 502, the first available model is converted into an onnx model.
And step 504, analyzing the onnx model based on TensorRT to obtain an Engine model.
Step 506, performing quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
The oonx, Open Neural Network Exchange format, is a standard for representing a deep learning model, and enables the model to be transferred between different frameworks. TensorRT is a high-performance inference C + + library and is specially applied to inference of edge equipment, the TensorRT can decompose and fuse the trained models, and the fused models have high integration degree. For example, after the convolution layer and the activation layer are fused, the inference speed can be improved. Before tensorRT engine inference, the picture channels need to be reordered and then reshape operated on the image matrix, becoming a one-dimensional matrix, since the input of the engine model must be one-dimensional and then input and output. In the embodiment, the trained model is subjected to marginalization processing through TensorRT, so that the magnitude of the model can be effectively reduced, and the reasoning speed is increased.
In one embodiment, as shown in fig. 6, after step 304, the method further includes:
step 601, extracting a ground wire area image in the first model training based on the position of the ground wire body.
Step 603, determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first labeling information.
Step 605, labeling the ground wire area image according to the abnormal position of the ground wire in the ground wire area image, and acquiring second model training data.
Step 607, training the initial YOLOV4-tiny model based on the second model training data to obtain the preset second recognition model.
Specifically, after the first model training data is obtained, the second model training data may be obtained based on further processing of the first model training data, and after the first model training data labels the position of the ground wire body in the historical image data, the historical image data may be trimmed based on the position of the ground wire body, so as to obtain a ground wire area image including the ground wire body. Meanwhile, the annotation in the model also comprises the abnormal position of the ground wire, so that the annotation on the image of the ground wire area can be updated based on the abnormal position of the ground wire to obtain second model training data; and then training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model. In a specific embodiment, the second model training data may be stored as a txt file in a YOLOV4 data format as a data set of the second model training data. In the concrete training process, before data is fed to a neural network, the sequence of a data set can be disordered, and then a Mosaic data enhancement strategy is adopted, so that the background and small targets of a detected object are enriched, and the data of four pictures can be calculated once when the Batch Normalization is calculated, so that the size of the mini-Batch does not need to be large, and the geometric enhancement comprises the following steps: random flipping (more horizontal flipping and less vertical flipping), random cropping (crop), stretching, and rotating. Wherein the color enhancement comprises: contrast enhancement, brightness enhancement, and more critical HSV spatial enhancement. Then, the image is normalized, the long side of the image is scaled to 416 pixels with the same size by adopting a self-adaptive image scaling algorithm, the short side is scaled by the same proportion, the part which is less than 416 takes the remainder of 64, then 0 is used for filling the short side of the image with the rest of length, and the label position information is subjected to the same transformation at the same time and then is sent to a network. Therefore, the data received by each epoch during training is different, and the robustness of the model can be increased. In this embodiment, the second model training data is obtained through the processing of the first model training, and the training data of the preset second recognition model can be obtained more efficiently, so that the training efficiency of the model is improved, and the reliability of the cascade model is ensured.
It should be understood that although the various steps in the flow charts of fig. 1-6 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 1-6 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps.
In one embodiment, as shown in fig. 7, there is provided a ground wire defect detecting apparatus, including:
the data acquisition module 702 is configured to acquire raw image data of a power tower region of a power grid.
The first identification module 704 is configured to determine position information of a ground wire in the original image data of the electric tower region by presetting a first identification model, where the preset first identification model is constructed based on a YOLOV4-tiny network structure by using a YOLOV4 algorithm.
The second identification module 706 is configured to input the position information of the ground lead to a preset second identification model, and determine the information of the defect of the ground lead in the original image data of the electric tower region, where the preset second identification model is constructed based on a YOLOV4-tiny network structure through a YOLOV4 algorithm.
And the risk warning module 708 is used for sending a risk warning according to the ground wire defect information.
In one embodiment, the first identification module 704 is specifically configured to: extracting a first characteristic diagram and a second characteristic diagram corresponding to the original image data of the electric tower area through a preset first identification model; performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map; processing the fusion feature map based on an attention mechanism to obtain a classification feature map; obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm; and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
In one embodiment, the first identification module 704 is specifically configured to: acquiring historical image data containing ground wire equipment and first marking information corresponding to the historical image data, wherein the first marking information is used for marking the body position of the ground wire and the abnormal position of the ground wire in the historical image data; marking historical image data through first marking information to obtain first model training data; and training the initial Yolov4-tiny model based on the first model training data to obtain a preset first recognition model.
In one embodiment, the first identification module 704 is specifically configured to: dividing the first model training data into training set data, test set data and verification set data; performing iterative training on the initial Yolov4-tiny model through training set data, test set data and verification set data, and performing model compression processing on the Yolov4-tiny model finished in each iteration in the iterative training process; when the loss of the training set data and the loss of the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold value, stopping iteration, and taking a YOLOV4-tiny model after current iteration training as a first available model; and performing marginalization processing on the first available model to obtain a preset first recognition model.
In one embodiment, the first identification module 704 is specifically configured to: converting the first available model to an onnx model; analyzing the onnx model based on TensorRT to obtain an Engine model; and carrying out quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
In one embodiment, the image segmentation processing module is further included to: extracting a ground wire area image in the first model training based on the position of the ground wire body; determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first marking information; marking the ground wire area image according to the abnormal position of the ground wire in the ground wire area image to obtain second model training data; and training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model.
For specific limitations of the ground wire defect detection apparatus, reference may be made to the above limitations on the ground wire defect detection method, which is not described herein again. All or part of each module in the ground wire defect detection device can be realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 8. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a method of ground wire defect detection. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 8 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program:
acquiring original image data of a power tower area of a power grid;
determining position information of a ground wire in original image data of an electric tower area through a preset first recognition model, and constructing the preset first recognition model based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, determining the defect information of the ground wires in the original image data of the electric tower area, and constructing the preset second recognition model based on a YOLOV4-tiny network structure through a YOLOV4 algorithm;
and sending a risk warning according to the fault information of the ground wire.
In one embodiment, the processor, when executing the computer program, further performs the steps of: extracting a first characteristic diagram and a second characteristic diagram corresponding to the original image data of the electric tower area through a preset first identification model; performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map; processing the fusion feature map based on an attention mechanism to obtain a classification feature map; obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm; and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
In one embodiment, the processor, when executing the computer program, further performs the steps of: acquiring historical image data containing ground wire equipment and first marking information corresponding to the historical image data, wherein the first marking information is used for marking the body position of the ground wire and the abnormal position of the ground wire in the historical image data; marking historical image data through first marking information to obtain first model training data; and training the initial Yolov4-tiny model based on the first model training data to obtain a preset first recognition model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: dividing the first model training data into training set data, test set data and verification set data; performing iterative training on the initial Yolov4-tiny model through training set data, test set data and verification set data, and performing model compression processing on the Yolov4-tiny model finished in each iteration in the iterative training process; when the loss of the training set data and the loss of the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold value, stopping iteration, and taking a YOLOV4-tiny model after current iteration training as a first available model; and performing marginalization processing on the first available model to obtain a preset first recognition model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: converting the first available model to an onnx model; analyzing the onnx model based on TensorRT to obtain an Engine model; and carrying out quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
In one embodiment, the processor, when executing the computer program, further performs the steps of: extracting a ground wire area image in the first model training based on the position of the ground wire body; determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first marking information; marking the ground wire area image according to the abnormal position of the ground wire in the ground wire area image to obtain second model training data; and training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of:
acquiring original image data of a power tower area of a power grid;
determining position information of a ground wire in original image data of an electric tower area through a preset first recognition model, and constructing the preset first recognition model based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, determining the defect information of the ground wires in the original image data of the electric tower area, and constructing the preset second recognition model based on a YOLOV4-tiny network structure through a YOLOV4 algorithm;
and sending a risk warning according to the fault information of the ground wire.
In one embodiment, the computer program when executed by the processor further performs the steps of: extracting a first characteristic diagram and a second characteristic diagram corresponding to the original image data of the electric tower area through a preset first identification model; performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map; processing the fusion feature map based on an attention mechanism to obtain a classification feature map; obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm; and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
In one embodiment, the computer program when executed by the processor further performs the steps of: acquiring historical image data containing ground wire equipment and first marking information corresponding to the historical image data, wherein the first marking information is used for marking the body position of the ground wire and the abnormal position of the ground wire in the historical image data; marking historical image data through first marking information to obtain first model training data; and training the initial Yolov4-tiny model based on the first model training data to obtain a preset first recognition model.
In one embodiment, the computer program when executed by the processor further performs the steps of: dividing the first model training data into training set data, test set data and verification set data; performing iterative training on the initial Yolov4-tiny model through training set data, test set data and verification set data, and performing model compression processing on the Yolov4-tiny model finished in each iteration in the iterative training process; when the loss of the training set data and the loss of the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold value, stopping iteration, and taking a YOLOV4-tiny model after current iteration training as a first available model; and performing marginalization processing on the first available model to obtain a preset first recognition model.
In one embodiment, the computer program when executed by the processor further performs the steps of: converting the first available model to an onnx model; analyzing the onnx model based on TensorRT to obtain an Engine model; and carrying out quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
In one embodiment, the computer program when executed by the processor further performs the steps of: extracting a ground wire area image in the first model training based on the position of the ground wire body; determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first marking information; marking the ground wire area image according to the abnormal position of the ground wire in the ground wire area image to obtain second model training data; and training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware related to instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile memory may include Read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, or the like. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above examples only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A method for detecting defects of a ground wire, which is implemented by an edge computing device, comprises the following steps:
acquiring original image data of a power tower area of a power grid;
determining position information of a conducting wire and a ground wire in the original image data of the electric tower area through a preset first identification model, wherein the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
inputting the position information of the ground wires into a preset second recognition model, and determining the defect information of the ground wires in the original image data of the electric tower area, wherein the preset second recognition model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and sending a risk warning according to the defect information of the ground wire.
2. The method according to claim 1, wherein the determining the position information of the ground wire in the raw image data of the electric tower area by presetting the first identification model comprises:
extracting a first feature map and a second feature map corresponding to the original image data of the electric tower region through the preset first identification model;
performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map;
processing the fusion feature map based on an attention mechanism to obtain a classification feature map;
obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm;
and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
3. The method according to claim 1, wherein before determining the position information of the ground wire in the raw image data of the electric tower region by presetting the first identification model, the method further comprises:
acquiring historical image data containing ground wire equipment and first marking information corresponding to the historical image data, wherein the first marking information is used for marking the body position of the ground wire and the abnormal position of the ground wire in the historical image data;
labeling the historical image data through the first labeling information to obtain first model training data;
and training the initial Yolov4-tiny model based on the first model training data to obtain a preset first recognition model.
4. The method of claim 3, wherein the training of the initial yoloov 4-tiny model based on the first model training data comprises:
dividing the first model training data into training set data, test set data and verification set data;
performing iterative training on the initial Yolov4-tiny model through the training set data, the test set data and the verification set data, and performing model compression processing on the Yolov4-tiny model which is finished each time of iteration in the iterative training process;
when the loss of the training set data and the loss of the test set data are in a stable state and the accuracy of the verification set data is higher than a preset accuracy threshold value, stopping iteration, and taking a YOLOV4-tiny model after current iteration training as a first available model;
and performing marginalization processing on the first available model to obtain a preset first recognition model.
5. The method according to claim 4, wherein the marginalizing the first available model to obtain the preset first recognition model comprises:
converting the first available model to an onnx model;
analyzing the onnx model based on TensorRT to obtain an Engine model;
and carrying out quantitative acceleration processing on the Engine model through TensorRT FP16 to obtain a preset first recognition model.
6. The method of claim 3, wherein after the labeling the historical image data by the first labeling information and obtaining the first model training data, further comprising:
extracting a ground wire area image in the first model training based on the position of the ground wire body;
determining the abnormal position of the ground wire in the image of the ground wire area based on the abnormal position of the ground wire in the first marking information;
marking the ground wire area image according to the abnormal position of the ground wire in the ground wire area image to obtain second model training data;
and training the initial Yolov4-tiny model based on the second model training data to obtain a preset second recognition model.
7. The utility model provides a lead ground wire defect detection device, is applied to marginal computing equipment, its characterized in that, the device includes:
the data acquisition module is used for acquiring original image data of an electric tower area of a power grid;
the first identification module is used for determining position information of a conducting wire in the original image data of the electric tower area through a preset first identification model, and the preset first identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
the second identification module is used for inputting the position information of the ground wires into a preset second identification model and determining the defect information of the ground wires in the original image data of the electric tower area, and the preset second identification model is constructed based on a Yolov4-tiny network structure through a Yolov4 algorithm;
and the risk warning module is used for sending risk warning according to the defect information of the ground wire.
8. The apparatus of claim 1, wherein the first identification module is specifically configured to: extracting a first feature map and a second feature map corresponding to the original image data of the electric tower region through the preset first identification model; performing upsampling processing on the first feature map, and fusing the upsampled first feature map and the second feature map to obtain a fused feature map; processing the fusion feature map based on an attention mechanism to obtain a classification feature map; obtaining a prediction frame of the classification characteristic diagram through a GIoU algorithm and an NMS algorithm; and determining the position information of the ground wire in the original image data of the electric tower area based on the prediction frame.
9. An edge computing device comprising a memory and a processor, the memory storing a computer program, wherein the processor when executing the computer program implements the steps of the method of any of claims 1 to 6.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111177578.0A CN113888514A (en) | 2021-10-09 | 2021-10-09 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111177578.0A CN113888514A (en) | 2021-10-09 | 2021-10-09 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113888514A true CN113888514A (en) | 2022-01-04 |
Family
ID=79005801
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111177578.0A Pending CN113888514A (en) | 2021-10-09 | 2021-10-09 | Method and device for detecting defects of ground wire, edge computing equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113888514A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114332634A (en) * | 2022-03-04 | 2022-04-12 | 浙江国遥地理信息技术有限公司 | Method and device for determining position of electric power tower at risk, electronic equipment and storage medium |
| CN114332659A (en) * | 2022-03-09 | 2022-04-12 | 南方电网数字电网研究院有限公司 | Power transmission line defect inspection method and device based on lightweight model issuing |
| CN114842363A (en) * | 2022-07-04 | 2022-08-02 | 南方电网科学研究院有限责任公司 | Identification method and system for key power equipment in digital twin platform area |
| CN114846998A (en) * | 2022-05-27 | 2022-08-05 | 云南农业大学 | Tomato picking method and system of binocular robot based on YOLOv4 algorithm |
| CN116128879A (en) * | 2023-04-17 | 2023-05-16 | 广东电网有限责任公司肇庆供电局 | Lightweight transmission line defect detection method and device |
-
2021
- 2021-10-09 CN CN202111177578.0A patent/CN113888514A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114332634A (en) * | 2022-03-04 | 2022-04-12 | 浙江国遥地理信息技术有限公司 | Method and device for determining position of electric power tower at risk, electronic equipment and storage medium |
| CN114332659A (en) * | 2022-03-09 | 2022-04-12 | 南方电网数字电网研究院有限公司 | Power transmission line defect inspection method and device based on lightweight model issuing |
| CN114846998A (en) * | 2022-05-27 | 2022-08-05 | 云南农业大学 | Tomato picking method and system of binocular robot based on YOLOv4 algorithm |
| CN114842363A (en) * | 2022-07-04 | 2022-08-02 | 南方电网科学研究院有限责任公司 | Identification method and system for key power equipment in digital twin platform area |
| CN114842363B (en) * | 2022-07-04 | 2022-10-18 | 南方电网科学研究院有限责任公司 | Identification method and system for key power equipment in digital twin platform area |
| CN116128879A (en) * | 2023-04-17 | 2023-05-16 | 广东电网有限责任公司肇庆供电局 | Lightweight transmission line defect detection method and device |
| CN116128879B (en) * | 2023-04-17 | 2023-08-01 | 广东电网有限责任公司肇庆供电局 | Lightweight transmission line defect detection method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113888514A (en) | Method and device for detecting defects of ground wire, edge computing equipment and storage medium | |
| CN115829999A (en) | Insulator defect detection model generation method, device, equipment and storage medium | |
| CN111680746B (en) | Vehicle damage detection model training, vehicle damage detection method, device, equipment and medium | |
| CN108229341A (en) | Sorting technique and device, electronic equipment, computer storage media, program | |
| WO2022160413A1 (en) | Electric power production anomaly monitoring method and apparatus, and computer device and storage medium | |
| CN113139543B (en) | Training method of target object detection model, target object detection method and equipment | |
| CN112419202B (en) | Automatic wild animal image recognition system based on big data and deep learning | |
| CN111368636A (en) | Object classification method and device, computer equipment and storage medium | |
| CA3193958A1 (en) | Processing images using self-attention based neural networks | |
| CN112184687B (en) | Road crack detection method based on capsule feature pyramid and storage medium | |
| CN112132216B (en) | Vehicle type recognition method and device, electronic equipment and storage medium | |
| CN113239869A (en) | Two-stage behavior identification method and system based on key frame sequence and behavior information | |
| CN113034514A (en) | Sky region segmentation method and device, computer equipment and storage medium | |
| US11232561B2 (en) | Capture and storage of magnified images | |
| CN111898693A (en) | Visibility classification model training method, visibility estimation method and device | |
| CN112508137A (en) | Transformer abnormality detection method and device, computer equipment and storage medium | |
| CN116468970A (en) | Model training method, image processing method, device, equipment and medium | |
| CN114241354A (en) | Warehouse personnel behavior identification method and device, computer equipment and storage medium | |
| CN114463613A (en) | Fault detection method and system based on residual error network and Faster R-CNN | |
| CN112528825A (en) | Station passenger recruitment service method based on image recognition | |
| CN111859370A (en) | Method, apparatus, electronic device and computer-readable storage medium for identifying service | |
| CN113837173A (en) | Target object detection method and device, computer equipment and storage medium | |
| CN115049836B (en) | Image segmentation method, device, equipment and storage medium | |
| CN117649358B (en) | Image processing method, device, equipment and storage medium | |
| CN113392739B (en) | Rolling bearing state monitoring method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |