CN116229380B - Method for identifying bird species related to bird-related faults of transformer substation - Google Patents
Method for identifying bird species related to bird-related faults of transformer substation Download PDFInfo
- Publication number
- CN116229380B CN116229380B CN202310518425.0A CN202310518425A CN116229380B CN 116229380 B CN116229380 B CN 116229380B CN 202310518425 A CN202310518425 A CN 202310518425A CN 116229380 B CN116229380 B CN 116229380B
- Authority
- CN
- China
- Prior art keywords
- module
- bird
- feature
- layer
- cbs
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Testing Of Short-Circuits, Discontinuities, Leakage, Or Incorrect Line Connections (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for identifying bird species related to a bird-involved fault of a transformer substation, which is based on a YOLOv8 target detection algorithm, introduces a triple attention mechanism, improves YOLOv8 by deformable convolution, constructs an improved YOLOv8 target detection model, and identifies bird species related to the bird-involved fault of the transformer substation through the trained improved YOLOv8 target detection model. The invention can provide reference for the identification research of the bird species related to the bird-related faults of the transformer substation. The invention enhances the feature extraction capability of the feature acquisition module, enhances the fusion effect of the feature fusion module and improves the loss function of the detection module, thereby finally enhancing the detection precision, generalization capability and robust performance of the model.
Description
Technical Field
The invention belongs to the technical field of transformer substation monitoring image data processing, and particularly relates to a method for identifying bird species related to bird-related faults of a transformer substation.
Background
The transformer station plays an important role in power transformation, concentration and distribution in a power system. In recent years, the fault events of transformer substation equipment caused by bird activities are increased year by year, and serious economic loss and social influence are brought. The transformer substation is widely distributed, most of the sites have less concentrated suburbs, and unique geographic conditions become the best choice for birds to camp. Bird trouble mainly includes trouble condition such as bird nest, bird's droppings, bird body short circuit, and bird nest material is in overcast and rainy wet weather or weather such as thunderbolt, leads to the short circuit accident of transformer substation inner conductor easily, and the bird droppings drops on insulators, sleeve pipe etc. reduce its insulating properties, leads to the flashover to cause short circuit tripping operation. Frequent activities such as bird nesting and defecation lead to gradual rise of the occurrence probability of bird damage in the transformer substation, and seriously threaten the safe and stable operation of the power grid. The existing bird pest control method utilizes the experience of patrol personnel and the bird pest control device to control bird pests, but the adaptability of birds to the bird pest control device in long-term operation leads to failure and resource waste. In actual work, the patrol personnel lack the relevant professional knowledge of birds, are difficult to identify the bird types in time, and are difficult to know the relevant activity rules, so that research on a tool for assisting the patrol personnel in identifying the birds is needed.
With the rapid development of deep learning, the deep learning has broken through the bottleneck of the traditional target detection algorithm and becomes a mainstream algorithm. There are two types of object detection algorithms that are currently popular. One type is a second order target detection algorithm, and the main idea is to generate a candidate region by using a selective search method and then to carry out regression classification in the candidate region. However, the main disadvantages of the second-order object detection algorithm are that the speed is slow and the calculation cost is high, and the second-order object detection algorithm requires more calculation and time to process the object in the image. In addition, the second order object detection algorithm also requires more parameters and features to be processed, which also increases the computational cost. In addition, since the second order object detection algorithm needs to process more information, the detection effect for a small object or a low resolution image may be affected. The other type is a first-order target detection algorithm, the algorithm simplifies the detection problem into a regression problem, the class probability and the position coordinates of the target can be directly obtained only by a convolutional neural network, the speed is higher than that of other algorithms, and the algorithm has excellent detection capability for small target detection. Therefore, the first-order target detection algorithm is required to be applied to the identification of the bird species related to the bird-related fault of the transformer substation, has the advantages of high detection speed, high precision and the like, can provide an effective bird identification tool for patrol personnel, and further improves the differential control effect of the bird-related fault of the transformer substation.
Disclosure of Invention
In view of the above, the invention aims to provide a method for identifying bird species related to a bird-related fault of a transformer substation, which is used for accurately identifying bird species related to the bird-related fault of the transformer substation, providing an effective bird identification tool for patrol personnel, and further improving the differential control effect of the bird-related fault of the transformer substation.
In order to achieve the above purpose, the present invention provides the following technical solutions: a method for identifying bird species related to a bird-involved fault of a transformer substation comprises the steps of introducing a triple attention mechanism (triplet attention) and deformable convolution (DCNv 2) to improve the YOLOv8 based on a YOLOv8 target detection algorithm, constructing an improved YOLOv8 target detection model, and identifying the bird species related to the bird-involved fault of the transformer substation through the trained improved YOLOv8 target detection model; the improved YOLOv8 target detection model comprises feature extractionThe device comprises a collection module, a feature fusion module and a detection module; the feature acquisition modules have 14 layers, namely DBS modules in layers 1, 2, 5, 8 and 11, C2F modules in layers 3, 6, 9 and 12, triple attention mechanisms in layers 4, 7, 10 and 13, SPPF modules in layer 14, and respectively extract features from layers 6, 9 and 14 as inputs of a feature fusion module, and outputs of layers 6, 9 and 14 are respectively extracted features F 1 Extracting features F 2 Extracting features F 3 The method comprises the steps of carrying out a first treatment on the surface of the The processing procedure of the feature fusion module is as follows: extracting features F 3 After upsampling and triple-attentiveness mechanism, and extracting feature F 2 Fusion to obtain enhanced feature F 4 Enhancement feature F 4 Through C2F 1 Post-module acquisition enhancement feature F 5 Enhancement feature F 5 After up-sampling and triple attention mechanism in turn, extracting characteristic F 1 Fusion to obtain enhanced feature F 6 Enhancement feature F 6 Through C2F 2 Post-module acquisition enhancement feature F 7 Enhancement feature F 7 Sequentially through DBS 1 Module and triple attention post-mechanism and enhanced feature F 5 Fusion to obtain enhanced feature F 8 Enhancement feature F 8 Through C2F 3 Post-module acquisition enhancement feature F 9 Enhancement feature F 9 Sequentially through DBS 2 Module and triple attention post-mechanism and extraction feature F 3 Fusion to obtain enhanced feature F 10 Enhancement feature F 10 Through C2F 4 Post-module acquisition enhancement feature F 11 Ultimately, feature F will be enhanced 7 Enhancement feature F 9 Enhancement feature F 11 As an output of the feature fusion module; DBS (DBS) 1 Module, DBS 2 The structure of the module is the same as that of a DBS module, and the DBS module is sequentially composed of a deformable convolution layer, a BN layer and a SiLU activation function; the detection module comprises three identical decoupling heads, which enhance the feature F 7 Enhancement feature F 9 Enhancement feature F 11 Respectively input into three decoupling heads.
Further preferably, C2F 1 Module, C2F 2 Module, C2F 3 Module, C2F 4 Structure of moduleAll are identical to the C2F module, the C2F module comprises 4 branches, the 1 st branch is composed of CBS 1 The 2 nd branch consists of CBS 1 The module and the separation module, the 3 rd branch is composed of CBS 1 The module, the separation module and the bottleneck module are formed, and the 4 th branch is formed by CBS 1 The module, the separation module and the two repeated bottleneck modules are formed, 4 branches are fused, and the module, the separation module and the two repeated bottleneck modules are formed by CBS 2 The module is used as output after feature integration, the separation modules passing through the 2 nd branch, the 3 rd branch and the 4 th branch are identical, and the bottleneck module in the 3 rd branch is also used as the first bottleneck module in the 4 th branch.
Further preferably, the SPPF module comprises 4 branches, the 1 st branch being defined by CBS 3 The module and the first maximum pooling layer, the 2 nd branch is composed of CBS 3 The module, the first maximum pooling layer and the second maximum pooling layer, the 3 rd branch is composed of CBS 3 The module, the first maximum pooling layer, the second maximum pooling layer and the third maximum pooling layer, the 4 th branch is composed of CBS 3 The module is composed of 4 branches which are fused and finally formed by CBS 4 The module performs feature integration and then serves as output.
Specifically, CBS 1 Module, CBS 2 Module, CBS 3 Module, CBS 4 The structure of the module is the same as that of a CBS module which is composed of a common convolution (conv), a BN layer and a SiLU activation function in sequence.
In particular, the Bottleneck module consists of a stack of 2 CBS modules and adds residual connections between the input and output, whereas the residual connections are removed when using the Bottleneck module (Bottleneck) in the feature fusion module.
Further preferably, the decoupling head has two identical branches, each consisting of 2 CBS modules and one common convolution (conv), one branch being responsible for prediction of the prediction block and the other branch being responsible for prediction of the category.
Further preferably, the decoupling head employs a Wise-IOU penalty function.
Further preferably, a substation bird-related fault bird species image database is constructed, and a training set and a test verification set are divided and are respectively used for improving training and verification of the YOLOv8 target detection model.
Further preferably, the method comprises the steps of collecting related bird species images shot by substation operation and maintenance personnel on site as an original image sample set, expanding the original image sample set to form a substation bird-related fault bird species image database, carrying out label making on the sample set of the substation bird-related fault bird species image database, dividing the sample set into a training set and a test verification set according to a certain proportion, and dividing the test verification set into a test set and a verification set.
Further preferably, an improved DCGAN algorithm is adopted to expand the original image sample set, and the improved DCGAN algorithm mainly comprises two modules of a generator G and a discriminator D; the generator G is provided with 10 layers in total, wherein the 1 st layer is a full connection layer, the 2 nd layer is an ECANet attention mechanism, the 3 rd to 8 th layers are convolution layers, the 9 th layer is a CA attention mechanism, the 10 th layer is a 256×256 convolution layer, the 3 th to 8 th convolution layers are added with BN layers and ReLU activation functions, and the activation functions of the 10 th convolution layer are Tanh functions; the arbiter D has 10 layers in total, the 1 st layer is 256×256 convolution layers, the 2 nd layer is CA attention mechanism, the 3 rd to 8 th layers are convolution layers, the 9 th layer is ECANet attention mechanism, the 10 th layer is full connection layer, the 3 th to 8 th convolution layers are added with BN layer and Leaky_ReLU activation function, and the full connection layer activation function of the 10 th layer is Sigmoid function.
Further preferably, the improved YOLOv8 target detection model is pre-trained by adopting the disclosed bird species image large-scale data set to obtain a pre-training weight, the pre-training weight is loaded, the improved YOLOv8 target detection model is retrained by utilizing the bird species image of the training set, and the Mosaic data is used for enhancement in the training process to obtain an optimal weight;
further preferably, loading the optimal weight, inputting the bird species image of the test set into an improved YOLOv8 target detection model for detection, and adopting a non-maximum value to inhibit NMS in a detection result to reject a prediction frame with the coincidence degree higher than a set value and the confidence score lower than the set value.
Compared with the prior art, the invention has the beneficial effects that: firstly, amplifying bird species images through an improved DCGAN algorithm, constructing a bird species image database related to bird-related faults of a transformer substation, and aiming at solving the problem of small image data samples; then introducing a triple attention mechanism, deformable convolution and a phase-IOU to improve a YOLOv8 target detection algorithm, enhancing the feature extraction capability of a feature acquisition module, enhancing the fusion effect of a feature fusion module and improving the loss function of a detection module, and finally enhancing the detection precision, generalization capability and robustness of the model; and finally, detecting the bird species image by using the trained improved YOLOv8 bird species target detection model, so that the accurate identification of the bird species related to the transformer substation bird-related fault can be realized. The technical scheme provided by the invention ensures the detection speed while maintaining higher accuracy, can provide an effective bird identification tool for patrol personnel, and further improves the differential control effect of the bird-involving faults of the transformer substation.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of a modified DCGAN algorithm generator G.
Fig. 3 is a schematic diagram of a modified DCGAN algorithm discriminator D.
FIG. 4 is a schematic diagram of an improved YOLOv8 target detection algorithm.
Fig. 5 is a schematic diagram of a C2F module.
Fig. 6 is a bottleneck module schematic.
Fig. 7 is a schematic diagram of an SPPF module.
Fig. 8 is a schematic diagram of a DBS module.
Fig. 9 is a schematic diagram of a CBS module.
Detailed Description
The present invention will now be further described with reference to the following examples, which are given by way of illustration only and are not to be construed as limiting the scope of the invention, since numerous insubstantial modifications and adaptations of the invention will now occur to those skilled in the art in light of the foregoing disclosure.
Referring to fig. 1, a method for identifying bird species related to bird-related faults of a transformer substation comprises the following steps:
s1: and constructing a bird-related fault bird species image database of the transformer substation. Collecting related bird species images shot by substation operation and maintenance personnel on site as an original image sample set, expanding the original image sample set by adopting an improved DCGAN algorithm, and carrying out label making on the expanded sample set and dividing the expanded sample set into a training set, a verification set and a test set according to a certain proportion;
in this embodiment, 5 birds including magpie, argan, vernonia nucifera, white head and swinery are selected as study objects, and according to the related bird species images shot by the operation and maintenance personnel on site, the number of each bird species image is 180, and 900 bird species images are counted as an original image sample set. The original image sample set was expanded using the modified DCGAN algorithm with 70 expansion of each bird species, thus a total of 250 images per bird species, for a total of 1250 bird species image samples. Then, the expanded sample set is processed by Labelimg label making software to be made into a data set in VOC format, and the data set is processed according to 9:1 into training set and test verification set (verification set + test set), and then the test verification set is divided into 9: the 1 scale is divided into a test set and a verification set, namely 1125 training sets, 113 test sets and 12 verification sets.
The improved DCGAN algorithm is constructed and mainly comprises two modules of a generator G and a discriminator D, as shown in fig. 2 and 3, and a CA attention mechanism and an ECANet attention mechanism are introduced into the two modules, wherein the specific structure is as follows: the generator G is provided with 10 layers in total, wherein the 1 st layer is a full connection layer, the 2 nd layer is an ECANet attention mechanism, the 3 rd to 8 th layers are convolution layers (4×4 convolution layers, 8×8 convolution layers, 16×016 convolution layers, 32×132 convolution layers, 64×264 convolution layers and 128×3128 convolution layers in sequence), the 9 th layer is a CA attention mechanism, the 10 th layer is a 256×256 convolution layer, BN layers and ReLU activation functions are added to the 3 rd to 8 th convolution layers, and the activation functions of the 10 th layer are Tanh functions; the discriminator D has 10 layers in total, layer 1 is 256×256 convolution layers, layer 2 is CA attention mechanism, layers 3-8 are convolution layers (128×128 convolution layers, 64×64 convolution layers, 32×32 convolution layers, 16×16 convolution layers, 8×8 convolution layers, 4×4 convolution layers in order), layer 9 is ECANet attention mechanism, layer 10 is fully connected layer, and layers 3-8The convolutional layer is added with BN layer and Leaky_ReLU activation function, and the full connection layer activation function of the 10 th layer is Sigmoid function. During training, 800 real bird images are used as network input for training, the training batch is set to 64, and the learning rates of the generator G and the discriminator D are all set to 1×10 -4 The training iteration number is 2500 rounds, and the parameter updating and optimizing are carried out on the discriminator D by adopting an SGDM optimizer and a random gradient descent method.
S2: based on the YOLOv8 target detection algorithm, a triple attention mechanism (triplet attention) and a deformable convolution (DCNv 2) are introduced to improve the YOLOv8, and an improved YOLOv8 target detection model is constructed. The improved YOLOv8 target detection model comprises a feature acquisition module, a feature fusion module and a detection module, as shown in fig. 4.
The characteristic acquisition module is provided with 14 layers, wherein the 1 st layer, the 2 nd layer, the 5 th layer, the 8 th layer and the 11 th layer are DBS modules, the 3 rd layer, the 6 th layer, the 9 th layer and the 12 th layer are C2F modules, the 4 th layer, the 7 th layer, the 10 th layer and the 13 th layer are triple attention mechanisms, the 14 th layer is an SPPF module, the DBS module can carry out convolution operation on input data, extract characteristics in the input data, carry out nonlinear transformation on output of a convolution layer by utilizing an activation function, enhance the expression capability of a network, and add a BN layer to normalize the output of the convolution layer, thereby improving the stability and the generalization capability of the network; the C2F module performs further feature extraction, fusion and enhanced feature expression on the input data, and the structure of the C2F module obtains richer gradient flow information while ensuring light weight; the SPPF module carries out pooling of different scales on the input feature images, and can extract feature information of different scales under the condition of not changing the size of the feature images, so that targets of different sizes are adapted, the size of the feature images can be reduced, the calculated amount is reduced, and meanwhile, the detection precision can be improved, so that the model can detect the targets more accurately. And extracting features from the 6 th layer, the 9 th layer and the 14 th layer respectively as inputs of the feature fusion module, and extracting features F from the 6 th layer, the 9 th layer and the 14 th layer respectively 1 Extracting features F 2 Extracting features F 3 。
The feature fusion module specifically extracts a feature F 3 After up-sampling and triple attention mechanism, and liftingTaking feature F 2 Fusion to obtain enhanced feature F 4 Enhancement feature F 4 Through C2F 1 Post-module acquisition enhancement feature F 5 Enhancement feature F 5 After up-sampling and triple attention mechanism in turn, extracting characteristic F 1 Fusion to obtain enhanced feature F 6 Enhancement feature F 6 Through C2F 2 Post-module acquisition enhancement feature F 7 Enhancement feature F 7 Sequentially through DBS 1 Module and triple attention post-mechanism and enhanced feature F 5 Fusion to obtain enhanced feature F 8 Enhancement feature F 8 Through C2F 3 Post-module acquisition enhancement feature F 9 Enhancement feature F 9 Sequentially through DBS 2 Module and triple attention post-mechanism and extraction feature F 3 Fusion to obtain enhanced feature F 10 Enhancement feature F 10 Through C2F 4 Post-module acquisition enhancement feature F 11 Ultimately, feature F will be enhanced 7 Enhancement feature F 9 Enhancement feature F 11 As an output of the feature fusion module.
The detection module specifically comprises: will enhance feature F 7 Enhancement feature F 9 Enhancement feature F 11 The method comprises the steps of respectively inputting the two branches into three identical decoupling heads, wherein each branch is composed of 2 CBS modules and a common convolution (conv), one branch is responsible for prediction of a prediction frame, the other branch is responsible for prediction of a variety, and a Wise-IOU loss function is utilized to replace the CIoU loss function of the original Yolov8, wherein the CBS modules can extract features in images, normalize input data and increase the nonlinearity of a network, so that the expression capacity and training speed of the network are improved.
C2F constructed in the present example 1 Module, C2F 2 Module, C2F 3 Module, C2F 4 The structure of the modules is the same as that of the C2F module, and the C2F module is formed by CBS 1 Module, CBS 2 A module, a Split module (Split) and a Bottleneck module (Bottleneck), wherein the Split module (Split) can Split the input data into different channels, each channel corresponding to a different feature, in order to enhance the networkThe expressive power enables the network to better capture the characteristics of different dimensions of the input data, each channel is sent to different convolution layers or other neural network layers for processing after passing through the separation module, so as to obtain a richer characteristic representation, the Bottleneck module (Bottleneck) can reduce the number of parameters and the calculation amount in the network, and can improve the precision and the generalization capability of the network, the C2F module comprises 4 branches, and the 1 st branch is formed by CBS (Conn.) as shown in FIG. 5 1 The 2 nd branch consists of CBS 1 The module and the separation module, the 3 rd branch is composed of CBS 1 The module, the separation module and the bottleneck module are formed, and the 4 th branch is formed by CBS 1 The module, the separation module and the two repeated bottleneck modules are formed, 4 branches are fused, and the module, the separation module and the two repeated bottleneck modules are formed by CBS 2 The module is used as output after feature integration, the separation modules passing through the 2 nd branch, the 3 rd branch and the 4 th branch are identical, and the bottleneck module in the 3 rd branch is also used as the first bottleneck module in the 4 th branch.
DBS 1 Module, DBS 2 The structure of the module is the same as the DBS module, which is composed of a deformable convolution layer (DCNv 2), BN layer and a SiLU activation function in that order, as shown in fig. 8.
The SPPF module includes 4 branches, as shown in fig. 7, branch 1 is defined by CBS 3 The module and the first maximum pooling layer (MaxPool), branch 2 consisting of CBS 3 The module, the first maximum pooling layer and the second maximum pooling layer, the 3 rd branch is composed of CBS 3 The module, the first maximum pooling layer, the second maximum pooling layer and the third maximum pooling layer, the 4 th branch is composed of CBS 3 The module is composed of 4 branches which are fused and finally formed by CBS 4 The module performs feature integration and then serves as output. CBS (cubic boron System) 1 Module, CBS 2 Module, CBS 3 Module, CBS 4 The structure of the module is the same as that of the CBS module, which is composed of a normal convolution (conv), BN layer, and a SiLU activation function in that order, as shown in fig. 9. The Bottleneck module (Bottleneck) consists of a stack of 2 CBS modules, as shown in fig. 6, and adds a residual connection between the input and output, while the Bottleneck module (Bottleneck) And removing the residual connection.
Compared with the common convolution, the deformable convolution (DCNv 2) is introduced, and the deformable convolution can enable the convolution kernel to have more flexible shapes and positions through the learnable deformation parameters, so that the shape and the positions of a target object can be more accurately adapted, and the performance of a deep learning model is further improved; the triple attention mechanism establishes the dependency relationship between the dimensions through rotation operation and residual transformation, compared with other attention mechanisms, a lighter weight computing structure is adopted, and the computing amount can be reduced and the interpretability is better while the higher precision is maintained; the Wise-IOU (dynamic non-monotonic focusing mechanism boundary frame loss) can evaluate the matching degree and positioning precision between the target frame and the real frame more accurately by introducing confidence information, and can obviously improve the performance of the model in certain specific scenes, so that the constructed improved YOLOv8 target detection algorithm has good detection capability and robustness;
s3: training an improved YOLOv8 target detection model. Pretraining the improved YOLOv8 target detection model by adopting the disclosed bird species image large-scale data set to obtain pretraining weights, loading the pretraining weights, re-training the improved YOLOv8 target detection model by utilizing the bird species image of the training set, and enhancing by using the Mosaic data in the training process to obtain optimal weights;
in the embodiment, a disclosed bird species image large-scale data set is adopted to pretrain an improved YOLOv8 target detection model, so as to obtain pretraining weights; based on the migration learning idea, loading pre-training weight as a basis, reusing bird species images of a training set to re-train an improved YOLOv8 target detection model, and setting the training round number to 150 in the training process by using Mosaic data enhancement at the first 10% of the training round number, the batch size to 16 and the initial learning rate to 1×10 -3 。
S4: and identifying the bird species related to the bird-related fault of the transformer substation through the trained improved YOLOv8 target detection model. Loading optimal weight, inputting bird species images of a test set into an improved YOLOv8 target detection model for detecting and verifying the effectiveness of the model, adopting a non-maximum value to inhibit NMS in a detection result to reject a prediction frame with the overlap ratio higher than a set value and the confidence score lower than the set value, wherein the overlap ratio set value is 0.3, the confidence set value is 0.5, and identifying bird species related to the transformer substation bird-related fault through the trained improved YOLOv8 target detection model.
This embodiment is accomplished in a software environment with a CPU of Intel Core i5-8300H, a main frequency of 2.90GHz, a GPU of Nvidia GeForce GTX 2060, cuda10, cudnn7.4.1.5, python, windows operating systems. This example references the average precision mean (mean average precision, mAP) to evaluate an improved Yolov8 bird species target detection model. Experimental results show that mAP values of 113 test pictures reach 97.522%, the effectiveness of the method provided by the embodiment is verified, an effective bird identification tool can be provided for patrol personnel, and the differential control effect of the transformer substation bird-related faults is further improved.
The foregoing description of the preferred embodiments of the invention is merely illustrative of and not limiting to the invention in its other forms, as modifications and equivalents may occur to others skilled in the art using the disclosure herein. However, any simple modification, equivalent variation and variation of the above embodiments according to the technical substance of the present invention still fall within the protection scope of the technical solution of the present invention.
Claims (9)
1. A method for identifying bird species related to a bird-related fault of a transformer substation is characterized by constructing a bird species image database related to the bird-related fault of the transformer substation, dividing a training set and a test verification set, and respectively improving training and verification of a YOLOv8 target detection model; based on a YOLOv8 target detection algorithm, introducing a triple attention mechanism and deformable convolution to improve YOLOv8, and constructing an improved YOLOv8 target detection model; pretraining the improved YOLOv8 target detection model with the disclosed large-scale data set of bird species images to obtain pretraining weights, loading pretraining weights, retraining the improved YOLOv8 target detection model with the bird species images of the training set, and enhancing with the use of Mosaic data during trainingObtaining optimal weights; performing substation bird-related fault related bird species identification through a trained improved YOLOv8 target detection model; the improved YOLOv8 target detection model comprises a feature acquisition module, a feature fusion module and a detection module; the feature acquisition modules have 14 layers, namely DBS modules in layers 1, 2, 5, 8 and 11, C2F modules in layers 3, 6, 9 and 12, triple attention mechanisms in layers 4, 7, 10 and 13, SPPF modules in layer 14, and respectively extract features from layers 6, 9 and 14 as inputs of a feature fusion module, and outputs of layers 6, 9 and 14 are respectively extracted features F 1 Extracting features F 2 Extracting features F 3 The method comprises the steps of carrying out a first treatment on the surface of the The processing procedure of the feature fusion module is as follows: extracting features F 3 After upsampling and triple-attentiveness mechanism, and extracting feature F 2 Fusion to obtain enhanced feature F 4 Enhancement feature F 4 Through C2F 1 Post-module acquisition enhancement feature F 5 Enhancement feature F 5 After up-sampling and triple attention mechanism in turn, extracting characteristic F 1 Fusion to obtain enhanced feature F 6 Enhancement feature F 6 Through C2F 2 Post-module acquisition enhancement feature F 7 Enhancement feature F 7 Sequentially through DBS 1 Module and triple attention post-mechanism and enhanced feature F 5 Fusion to obtain enhanced feature F 8 Enhancement feature F 8 Through C2F 3 Post-module acquisition enhancement feature F 9 Enhancement feature F 9 Sequentially through DBS 2 Module and triple attention post-mechanism and extraction feature F 3 Fusion to obtain enhanced feature F 10 Enhancement feature F 10 Through C2F 4 Post-module acquisition enhancement feature F 11 Ultimately, feature F will be enhanced 7 Enhancement feature F 9 Enhancement feature F 11 As an output of the feature fusion module; DBS (DBS) 1 Module, DBS 2 The structure of the module is the same as that of a DBS module, and the DBS module is sequentially composed of a deformable convolution layer, a BN layer and a SiLU activation function; the detection module comprises three identical decoupling heads, which enhance the feature F 7 Enhancement feature F 9 Enhancement feature F 11 Respectively input to threeIn the decoupling head.
2. The method for identifying bird species associated with bird-related faults of transformer substation according to claim 1, wherein the method is characterized by C2F 1 Module, C2F 2 Module, C2F 3 Module, C2F 4 The structure of the modules is the same as that of the C2F module, the C2F module comprises 4 branches, and the 1 st branch is formed by CBS 1 The 2 nd branch consists of CBS 1 The module and the separation module, the 3 rd branch is composed of CBS 1 The module, the separation module and the bottleneck module are formed, and the 4 th branch is formed by CBS 1 The module, the separation module and the two repeated bottleneck modules are formed, 4 branches are fused, and the module, the separation module and the two repeated bottleneck modules are formed by CBS 2 The module is used as output after feature integration, the separation modules passing through the 2 nd branch, the 3 rd branch and the 4 th branch are identical, and the bottleneck module in the 3 rd branch is also used as the first bottleneck module in the 4 th branch.
3. The method for identifying bird species associated with a bird-involved fault in a transformer substation according to claim 2, wherein the SPPF module comprises 4 branches, the 1 st branch being defined by CBS 3 The module and the first maximum pooling layer, the 2 nd branch is composed of CBS 3 The module, the first maximum pooling layer and the second maximum pooling layer, the 3 rd branch is composed of CBS 3 The module, the first maximum pooling layer, the second maximum pooling layer and the third maximum pooling layer, the 4 th branch is composed of CBS 3 The module is composed of 4 branches which are fused and finally formed by CBS 4 The module performs feature integration and then serves as output.
4. The method for identifying bird species associated with a bird-involved fault in a substation according to claim 3, wherein in particular CBS 1 Module, CBS 2 Module, CBS 3 Module, CBS 4 The structure of the module is the same as that of a CBS module, and the CBS module is sequentially composed of a common convolution, a BN layer and a SiLU activation function.
5. The method for identifying bird species associated with a bird-involved fault in a transformer substation according to claim 2, wherein the bottleneck module is composed of a stack of 2 CBS modules, and a residual connection is added between the input and the output, and the residual connection is removed when the bottleneck module is used in the feature fusion module.
6. The method for identifying bird species related to bird-related faults of a transformer substation according to claim 1, wherein the decoupling head is provided with two identical branches, each of which consists of 2 CBS modules and one common convolution, one branch is responsible for the prediction of a prediction frame, and the other branch is responsible for the prediction of a species; the decoupling head adopts a Wise-IOU loss function.
7. The method for identifying the bird species related to the bird fault of the transformer substation according to claim 1 is characterized by collecting related bird species images shot by operation staff of the transformer substation on site as an original image sample set, expanding the original image sample set to form a bird species image database related to the bird fault of the transformer substation, labeling the sample set of the bird species image database related to the bird fault of the transformer substation and dividing the sample set into a training set and a test verification set according to proportion, and dividing the test verification set into a test set and a verification set.
8. The method for identifying bird species related to bird-related faults of a transformer substation according to claim 7, wherein an original image sample set is expanded by adopting an improved DCGAN algorithm, and the improved DCGAN algorithm consists of two modules of a generator G and a discriminator D; the generator G is provided with 10 layers in total, wherein the 1 st layer is a full connection layer, the 2 nd layer is an ECANet attention mechanism, the 3 rd to 8 th layers are convolution layers, the 9 th layer is a CA attention mechanism, the 10 th layer is a 256×256 convolution layer, the 3 th to 8 th convolution layers are added with BN layers and ReLU activation functions, and the activation functions of the 10 th convolution layer are Tanh functions; the arbiter D has 10 layers in total, the 1 st layer is 256×256 convolution layers, the 2 nd layer is CA attention mechanism, the 3 rd to 8 th layers are convolution layers, the 9 th layer is ECANet attention mechanism, the 10 th layer is full connection layer, the 3 th to 8 th convolution layers are added with BN layer and Leaky_ReLU activation function, and the full connection layer activation function of the 10 th layer is Sigmoid function.
9. The method for identifying bird species related to bird-related faults of the transformer substation according to claim 1 is characterized in that optimal weights are loaded, the image of the bird species collected to be tested is input into an improved YOLOv8 target detection model for detection, and a prediction frame with the coincidence degree higher than a set value and the confidence score lower than the set value is removed by adopting a non-maximum value inhibition NMS in a detection result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310518425.0A CN116229380B (en) | 2023-05-10 | 2023-05-10 | Method for identifying bird species related to bird-related faults of transformer substation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310518425.0A CN116229380B (en) | 2023-05-10 | 2023-05-10 | Method for identifying bird species related to bird-related faults of transformer substation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116229380A CN116229380A (en) | 2023-06-06 |
| CN116229380B true CN116229380B (en) | 2023-09-12 |
Family
ID=86589635
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310518425.0A Active CN116229380B (en) | 2023-05-10 | 2023-05-10 | Method for identifying bird species related to bird-related faults of transformer substation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116229380B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116994287B (en) * | 2023-07-04 | 2024-05-24 | 北京市农林科学院 | Animal counting method and device and animal counting equipment |
| CN116935473A (en) * | 2023-07-28 | 2023-10-24 | 山东智和创信息技术有限公司 | Real-time detection method and system for wearing safety helmet based on improved YOLO v7 under complex background |
| CN117113066B (en) * | 2023-10-25 | 2024-03-29 | 南昌大学 | Transmission line insulator defect detection method based on computer vision |
| CN117994253B (en) * | 2024-04-03 | 2024-06-11 | 国网山东省电力公司东营供电公司 | High-voltage distribution line ground fault identification method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115240012A (en) * | 2022-08-19 | 2022-10-25 | 国网四川省电力公司电力科学研究院 | Bird detection method and bird detection system for power transmission line based on improved YOLOv5 |
| CN115620107A (en) * | 2022-11-07 | 2023-01-17 | 国网江西省电力有限公司电力科学研究院 | Transformer substation bird-involved fault related bird species identification method based on deep learning |
| CN115862073A (en) * | 2023-02-27 | 2023-03-28 | 国网江西省电力有限公司电力科学研究院 | Transformer substation harmful bird species target detection and identification method based on machine vision |
| CN115937655A (en) * | 2023-02-24 | 2023-04-07 | 城云科技（中国）有限公司 | Target detection model of multi-order feature interaction, and construction method, device and application thereof |
| CN115984172A (en) * | 2022-11-29 | 2023-04-18 | 上海师范大学 | Small target detection method based on enhanced feature extraction |
| CN116052207A (en) * | 2022-12-07 | 2023-05-02 | 国网四川省电力公司阿坝供电公司 | Bird identification method, system and medium based on improved YOLOv7 |
| CN116071667A (en) * | 2023-04-07 | 2023-05-05 | 北京理工大学 | Method and system for detecting abnormal aircraft targets in specified area based on historical data |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10909424B2 (en) * | 2018-10-13 | 2021-02-02 | Applied Research, LLC | Method and system for object tracking and recognition using low power compressive sensing camera in real-time applications |
| US11109586B2 (en) * | 2019-11-13 | 2021-09-07 | Bird Control Group, Bv | System and methods for automated wildlife detection, monitoring and control |
-
2023
- 2023-05-10 CN CN202310518425.0A patent/CN116229380B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115240012A (en) * | 2022-08-19 | 2022-10-25 | 国网四川省电力公司电力科学研究院 | Bird detection method and bird detection system for power transmission line based on improved YOLOv5 |
| CN115620107A (en) * | 2022-11-07 | 2023-01-17 | 国网江西省电力有限公司电力科学研究院 | Transformer substation bird-involved fault related bird species identification method based on deep learning |
| CN115984172A (en) * | 2022-11-29 | 2023-04-18 | 上海师范大学 | Small target detection method based on enhanced feature extraction |
| CN116052207A (en) * | 2022-12-07 | 2023-05-02 | 国网四川省电力公司阿坝供电公司 | Bird identification method, system and medium based on improved YOLOv7 |
| CN115937655A (en) * | 2023-02-24 | 2023-04-07 | 城云科技（中国）有限公司 | Target detection model of multi-order feature interaction, and construction method, device and application thereof |
| CN115862073A (en) * | 2023-02-27 | 2023-03-28 | 国网江西省电力有限公司电力科学研究院 | Transformer substation harmful bird species target detection and identification method based on machine vision |
| CN116071667A (en) * | 2023-04-07 | 2023-05-05 | 北京理工大学 | Method and system for detecting abnormal aircraft targets in specified area based on historical data |
Non-Patent Citations (1)
| Title |
|---|
| 邱志斌等.基于深度迁移学习的输电线路涉鸟故障危害鸟种图像识别.《高电压技术》.2021,第47卷(第11期),全文. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116229380A (en) | 2023-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN116229380B (en) | Method for identifying bird species related to bird-related faults of transformer substation | |
| CN109873610B (en) | Photovoltaic array fault diagnosis method based on IV characteristic and depth residual error network | |
| CN111680706B (en) | Dual-channel output contour detection method based on coding and decoding structure | |
| CN114092832B (en) | High-resolution remote sensing image classification method based on parallel hybrid convolutional network | |
| CN107564025A (en) | A kind of power equipment infrared image semantic segmentation method based on deep neural network | |
| CN114067368B (en) | Power grid harmful bird species classification and identification method based on deep convolution characteristics | |
| CN109635928A (en) | A kind of voltage sag reason recognition methods based on deep learning Model Fusion | |
| CN110427846A (en) | It is a kind of using convolutional neural networks to the face identification method of uneven small sample | |
| CN109934222A (en) | A kind of insulator chain self-destruction recognition methods based on transfer learning | |
| CN106920007A (en) | PM based on second order Self-organized Fuzzy Neural Network2.5Intelligent Forecasting | |
| CN110969194B (en) | Cable early fault positioning method based on improved convolutional neural network | |
| CN111652835A (en) | Method for detecting insulator loss of power transmission line based on deep learning and clustering | |
| CN111369057A (en) | Air quality prediction optimization method and system based on deep learning | |
| CN115994325B (en) | Fan icing power generation data enhancement method based on TimeGAN deep learning method | |
| CN113255661B (en) | Bird species image identification method related to bird-involved fault of power transmission line | |
| CN114842208A (en) | Power grid harmful bird species target detection method based on deep learning | |
| CN112215279B (en) | Power grid fault diagnosis method based on immune RBF neural network | |
| CN116612098B (en) | Insulator RTV spraying quality evaluation method and device based on image processing | |
| CN113589100A (en) | Abnormal discharge identification method based on bidirectional long-short term memory and attention mechanism | |
| CN114519293A (en) | Cable body fault identification method based on hand sample machine learning model | |
| CN110726813B (en) | Electronic nose prediction method based on double-layer integrated neural network | |
| CN116611580A (en) | Ocean red tide prediction method based on multi-source data and deep learning | |
| CN116561692A (en) | Dynamic update real-time measurement data detection method | |
| CN110543675A (en) | Power transmission line fault identification method | |
| Colak et al. | Automatic sunspot classification for real-time forecasting of solar activities |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |