CN116862867A - Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN - Google Patents
Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN Download PDFInfo
- Publication number
- CN116862867A CN116862867A CN202310826080.5A CN202310826080A CN116862867A CN 116862867 A CN116862867 A CN 116862867A CN 202310826080 A CN202310826080 A CN 202310826080A CN 116862867 A CN116862867 A CN 116862867A
- Authority
- CN
- China
- Prior art keywords
- substation equipment
- training
- anogan
- network
- classification
- 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
- 230000007547 defect Effects 0.000 title claims abstract description 82
- 238000001514 detection method Methods 0.000 title claims abstract description 58
- 230000000007 visual effect Effects 0.000 title claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 133
- 230000002950 deficient Effects 0.000 claims abstract description 68
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000005457 optimization Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 abstract description 6
- 230000002159 abnormal effect Effects 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 8
- 230000009466 transformation Effects 0.000 description 8
- 238000007689 inspection Methods 0.000 description 7
- 238000004422 calculation algorithm Methods 0.000 description 6
- 238000004590 computer program Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 4
- 238000012544 monitoring process Methods 0.000 description 4
- 238000013145 classification model Methods 0.000 description 3
- 238000003745 diagnosis Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000005856 abnormality Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 239000000427 antigen Substances 0.000 description 1
- 102000036639 antigens Human genes 0.000 description 1
- 108091007433 antigens Proteins 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
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
- 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
- G06N3/0475—Generative 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
- G06N3/094—Adversarial learning
-
- 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
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/761—Proximity, similarity or dissimilarity measures
-
- 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Multimedia (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)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a visual defect detection method for small sample substation equipment based on improved AnoGAN, and belongs to the technical field of machine learning; the method comprises the following steps: acquiring an image of training substation equipment; training an AnoGAN countermeasure generation network according to the normal substation equipment image; inputting the images of the defective substation equipment into an AnoGAN countermeasure generation network after training is completed; inputting the preprocessed image of the defective substation equipment into a pre-trained similarity classification network for classification; inputting the preprocessed defective substation equipment image into a pre-trained convolution classification network for classification; and obtaining a defect detection result according to the similarity classification result and the convolution classification result. The invention also provides a small sample substation equipment visual defect detection system based on the improved AnoGAN. The training network trains aiming at normal data, so that the generation effect of the training network can distinguish all abnormal state equipment, and the training network has higher universality in practical application.
Description
Technical Field
The invention relates to the technical field of machine learning, in particular to a small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN.
Background
With the development of artificial intelligence technology, the intelligent inspection technology is increasingly applied to the transformer substation, and the intelligent inspection technology has the main function of replacing manual work to finish daily inspection of the unattended transformer substation. The effective way for realizing intelligent inspection of the transformer substation is to combine the existing video processing technology with the image recognition technology, so as to realize real-time monitoring of defects of transformer substation equipment.
The prior art generally adopts a general neural network to train a detection model for defect detection. However, due to the characteristics of the electric power field, large quantities of real defect data are difficult to obtain, real scenes cannot be covered, the training model is subjected to fitting, and the omission ratio is high in practical application. By performing defect detection on the countermeasure neural network pair, a data set can be automatically generated and co-training is performed with the detection network, the robustness of the detection network is improved while the accuracy of the generated data is improved, and the problems that the current defect data is insufficient and the detection rate of the detection network is high can be well solved.
The current visual defect detection method is mostly based on a traditional detection model, a YOLO series, a mask-CNN and the like, and the algorithm is not designed for defect detection, and only takes the defect as a detection type. Therefore, the trained model has higher omission rate for defect types outside the training set, and the model needs to be iterated continuously to adapt to different application environments.
In addition, due to the characteristics of the electric power field, real defect data are difficult to obtain in actual training, so model optimization is required for small samples, in the existing method, an antigen generation network (GAN) is used as an auxiliary tool for generating an artificial data set to expand the data set, but in the actual application situation, the gap between data generated by the GAN and an original image is large, and the positive influence on the model is small. The existing GAN-CNN method can solve the problem of insufficient data sets to a certain extent, but the GAN algorithm still cannot generate algorithms beyond the types of the data sets, a network trained by the method lacks robustness, and the detection omission is easy to occur aiming at the defects of fewer data sets in practical application.
The prior art scheme is as follows:
1. the invention discloses a switch cabinet partial discharge fault mode identification method based on GAN-CNN, which comprises the following steps of: CN202211659218
The invention discloses a switch cabinet partial discharge fault mode identification method based on GAN-CNN, which comprises the following steps: s1, collecting partial discharge PRPD map data of a transformer substation switch cabinet; s2, combining priori knowledge and a real PRPD map, and generating a simulated PRPD partial discharge defect map which can be used for classification model learning based on a GAN network; s3, randomly extracting part of real PRPD patterns to serve as a test set, scrambling the simulated PRPD patterns and the residual real PRPD patterns, and inputting the scrambled simulated PRPD patterns and the residual real PRPD patterns serving as a training set into a CNN (computer numerical network) for training to obtain a switch cabinet partial discharge fault mode identification model; and S4, carrying out partial discharge mode identification on the real-time PRPD map by using the trained model, and realizing situation awareness and anomaly monitoring of the transformer substation switch cabinet. According to the invention, feature analysis can be carried out on massive PRPD patterns, and the accuracy of the recognition model is improved while the problem of scarcity of negative samples in the historical training set is effectively solved by carrying out deep mining on PRPD data.
2. A high-precision identification method for defects of a transformer device based on a mask-CNN algorithm framework is disclosed in the patent application number: CN202211203755
The invention discloses a high-precision identification method for defects of a power transformation device based on a mask-CNN algorithm framework, which comprises the following steps: collecting a defect image data set of the transformer equipment; summarizing the target power transformation equipment and defect types of the target power transformation equipment according to the actual condition of the power transformation equipment site to obtain a training data set; the GAN network is utilized to increase the defect image data set, the defect of insufficient training data set is filled, and an expanded defect map sample set of the power transformation equipment is obtained; and performing network training learning on the expanded substation equipment defect map sample set by using the optimized GFPN network model of the mask-CNN to realize identification of the existing substation equipment defects. According to the invention, the defect characteristics of the power transformation equipment are extracted through a depth vision technology, and when the defect of the power transformation equipment is found in the monitoring or inspection process, an identification and classification result is timely made; the method provided by the invention not only can greatly reduce the workload of inspection of the power transformation equipment, but also can improve the inspection efficiency and ensure the safety of power operation.
3. Substation equipment fault diagnosis method based on integrated depth generation model is disclosed in patent application number: CN202211014277
The invention discloses a substation equipment fault diagnosis method based on an integrated depth generation model, which comprises the following steps: collecting substation equipment operation data, preprocessing the data, and constructing a sample training data set and a random potential data set; constructing an AMBI-GAN integrated depth generation model based on a two-way long-short-term memory network and an attention mechanism; training an AMBI-GAN integrated depth generation model; inputting test data into the trained model, and calculating the total loss Ltest of the model; obtaining an identification score through 1-Ltest, and judging abnormality when the identification score exceeds a preset value; the invention has the advantages that: and a small amount of marked data is fully utilized to further improve the fault diagnosis performance.
Disclosure of Invention
The invention aims to provide an efficient visual defect detection method for small sample substation equipment based on improved AnoGAN.
In order to solve the technical problems, the invention provides a small sample transformer substation equipment visual defect detection method based on improved AnoGAN, which comprises the following steps:
acquiring an image of training substation equipment; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
training an AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
inputting the images of the defective substation equipment into an AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
inputting the preprocessed image of the defective substation equipment into a pre-trained similarity classification network for classification, and outputting a similarity classification result;
inputting the preprocessed defective substation equipment image into a pre-trained convolution classification network for classification, and outputting a convolution classification result;
and obtaining a defect detection result according to the similarity classification result and the convolution classification result.
Preferably, the AnoGAN challenge-generating network includes a generator and a discriminator, and the AnoGAN challenge-generating network is trained according to the normal substation equipment image, so as to obtain the AnoGAN challenge-generating network after training is completed, and the method specifically includes the following steps:
inputting the normal substation equipment image into a generator of an AnoGAN countermeasure generation network to obtain a target substation equipment image;
inputting the normal substation equipment image and the target substation equipment image into a discriminator of an AnoGAN countermeasure generation network to obtain a discrimination result;
and optimizing a generator according to the distinguishing result to obtain the AnoGAN countermeasure generation network after training is completed.
Preferably, the normal substation equipment image and the target substation equipment image are input into a discriminator of the AnoGAN countermeasure generation network to obtain a discrimination result, and the method specifically comprises the following steps:
calculating peak signal-to-noise ratio PSNR of each normal substation equipment image and each target substation equipment image;
and calculating to obtain a PSNR average value according to the peak signal-to-noise ratio PSNR of all the normal substation equipment images and the target substation equipment images, and taking the PSNR average value as a distinguishing result.
Preferably, the generator is optimized according to the discrimination result to obtain the AnoGAN countermeasure generation network after training is completed, and the method specifically comprises the following steps:
optimizing network parameters of the generator according to the PSNR average value to obtain an optimized generator;
judging whether the PSNR average value is larger than a first PSNR threshold value or not;
if the PSNR average value is less than or equal to the first PSNR threshold value, retraining the AnoGAN countermeasure generation network;
and ending the optimization until the PSNR average value is larger than the first PSNR threshold value, and obtaining the AnoGAN countermeasure generation network after the training is completed according to the optimized generator.
Preferably, the training method of the similarity classification network after pre-training is as follows:
inputting the training set into a similarity classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image for training and the defective substation equipment image for training;
setting a second PSNR threshold according to the peak signal-to-noise ratio PSNR of the training normal substation equipment image and the training defective substation equipment image;
and obtaining the similarity classification network after pre-training according to the second PSNR threshold value.
Preferably, the second PSNR threshold is less than or equal to 99% of peak signal-to-noise ratio PSNR of the training normal substation equipment image and the corresponding training defective substation equipment image in the training set.
Preferably, the preprocessed defective substation equipment image is input into a pre-trained similarity classification network for classification, and a similarity classification result is output, and the method specifically comprises the following steps:
inputting the normal substation equipment image and the preprocessed defective substation equipment image into a pre-trained similarity classification network;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image and the preprocessed defective substation equipment image;
judging whether the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than a second PSNR threshold value or not;
if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than or equal to a second PSNR threshold value, judging that the to-be-detected substation equipment image is normal;
and if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is smaller than the second PSNR threshold, judging that the to-be-detected substation equipment image has corresponding defects.
Preferably, the classification method of the pre-trained convolutional classification network comprises the following steps:
inputting the training set into a convolutional classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
and training the convolutional classification network according to the normal substation equipment image and the training defective substation equipment image based on the supervised learning method to obtain the pre-trained convolutional classification network.
Preferably, a defect detection result is obtained according to a similarity classification result and a convolution classification result, and the method specifically comprises the following steps:
and taking the intersection of the similarity classification result and the convolution classification result as a defect detection result.
The invention also provides a small sample substation equipment visual defect detection system based on the improved AnoGAN, which comprises the following steps:
the acquisition module is used for acquiring the training substation equipment image; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
the AnoGAN countermeasure generation network training module is used for training the AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
the preprocessing module is used for inputting the images of the defective substation equipment into the AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
the similarity classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained similarity classification network for classification and outputting a similarity classification result;
the convolution classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained convolution classification network for classification and outputting convolution classification results;
and the defect detection module is used for obtaining a defect detection result according to the similarity classification result and the convolution classification result.
Compared with the prior art, the invention has the beneficial effects that:
in the invention, the AnoGAN countermeasure generation network is used for training non-fault data, so that the AnoGAN countermeasure generation network has natural recognition capability for all kinds of fault data, and judgment on the generation effect of the AnoGAN countermeasure generation network is added for a plurality of times in a training process so as to ensure the similarity of the generated data and the real data. Secondly, in the invention, the AnoGAN countermeasure generation network is taken as a part of the discrimination model and participates in the subsequent model joint debugging process, thereby being beneficial to improving the recognition precision.
In the invention, the AnoGAN countermeasure generation network does not participate in data enhancement, and on the premise of ensuring that the similarity between the generated data and the original data is higher, the GAN data only participates in the preprocessing of the image, thereby ensuring the effect of the subsequent model training data. Secondly, because the training network only trains the normal data in the method, the generation effect of the training network can distinguish all abnormal state equipment, and the training network has higher universality in practical application.
In the invention, the discriminator of the AnoGAN countermeasure generation network is optimized aiming at the image field, the discriminator compresses the picture size by adopting 4-weight down_block, and the model is optimized by adopting 1x1 convolution instead of the prior fully-connected network. Secondly, in the invention, PSNR is used for judging the image similarity, the parameter is not used in a training generation network, the fitting problem does not exist, the network external convolution classification network uses the result jointly output by the classification network and the similarity network as final judgment, and the accuracy is higher.
Drawings
The following describes the embodiments of the present invention in further detail with reference to the accompanying drawings.
FIG. 1 is a schematic flow chart of a visual defect detection method in example 1;
FIG. 2 is a schematic diagram of a network architecture of a discriminator;
fig. 3 is a schematic flow chart of a method for detecting visual defects of small-sample substation equipment based on improved AnoGAN.
Detailed Description
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. The present invention may be embodied in many other forms than those herein described, and those skilled in the art will readily appreciate that the present invention may be similarly embodied without departing from the spirit or essential characteristics thereof, and therefore the present invention is not limited to the specific embodiments disclosed below.
The terminology used in the one or more embodiments of the specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of the specification. As used in this specification, one or more embodiments and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and/or" as used in one or more embodiments of the present specification refers to and encompasses any or all possible combinations of one or more of the associated listed items.
It should be understood that, although the terms first, second, etc. may be used in one or more embodiments of this specification to describe various information, these information should not be limited by these terms. These terms are only used to distinguish one type of information from another. For example, a first may also be referred to as a second, and similarly, a second may also be referred to as a first, without departing from the scope of one or more embodiments of the present description. The word "if" as used herein may be interpreted as "at … …" or "at … …" or "responsive to a determination", depending on the context.
The invention is described in further detail below with reference to the attached drawing figures:
as shown in fig. 3, the invention provides a small sample substation equipment visual defect detection method based on improved AnoGAN, which comprises the following steps:
acquiring an image of training substation equipment; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
training an AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
inputting the images of the defective substation equipment into an AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
inputting the preprocessed image of the defective substation equipment into a pre-trained similarity classification network for classification, and outputting a similarity classification result;
inputting the preprocessed defective substation equipment image into a pre-trained convolution classification network for classification, and outputting a convolution classification result;
and obtaining a defect detection result according to the similarity classification result and the convolution classification result.
Preferably, the AnoGAN challenge-generating network includes a generator and a discriminator, and the AnoGAN challenge-generating network is trained according to the normal substation equipment image, so as to obtain the AnoGAN challenge-generating network after training is completed, and the method specifically includes the following steps:
inputting the normal substation equipment image into a generator of an AnoGAN countermeasure generation network to obtain a target substation equipment image;
inputting the normal substation equipment image and the target substation equipment image into a discriminator of an AnoGAN countermeasure generation network to obtain a discrimination result;
and optimizing a generator according to the distinguishing result to obtain the AnoGAN countermeasure generation network after training is completed.
Preferably, the normal substation equipment image and the target substation equipment image are input into a discriminator of the AnoGAN countermeasure generation network to obtain a discrimination result, and the method specifically comprises the following steps:
calculating peak signal-to-noise ratio PSNR of each normal substation equipment image and each target substation equipment image;
and calculating to obtain a PSNR average value according to the peak signal-to-noise ratio PSNR of all the normal substation equipment images and the target substation equipment images, and taking the PSNR average value as a distinguishing result.
Preferably, the generator is optimized according to the discrimination result to obtain the AnoGAN countermeasure generation network after training is completed, and the method specifically comprises the following steps:
optimizing network parameters of the generator according to the PSNR average value to obtain an optimized generator;
judging whether the PSNR average value is larger than a first PSNR threshold value or not;
if the PSNR average value is less than or equal to the first PSNR threshold value, retraining the AnoGAN countermeasure generation network;
and ending the optimization until the PSNR average value is larger than the first PSNR threshold value, and obtaining the AnoGAN countermeasure generation network after the training is completed according to the optimized generator.
Preferably, the training method of the similarity classification network after pre-training is as follows:
inputting the training set into a similarity classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image for training and the defective substation equipment image for training;
setting a second PSNR threshold according to the peak signal-to-noise ratio PSNR of the training normal substation equipment image and the training defective substation equipment image;
and obtaining the similarity classification network after pre-training according to the second PSNR threshold value.
Preferably, the second PSNR threshold is less than or equal to 99% of peak signal-to-noise ratio PSNR of the training normal substation equipment image and the corresponding training defective substation equipment image in the training set.
Preferably, the preprocessed defective substation equipment image is input into a pre-trained similarity classification network for classification, and a similarity classification result is output, and the method specifically comprises the following steps:
inputting the normal substation equipment image and the preprocessed defective substation equipment image into a pre-trained similarity classification network;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image and the preprocessed defective substation equipment image;
judging whether the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than a second PSNR threshold value or not;
if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than or equal to a second PSNR threshold value, judging that the to-be-detected substation equipment image is normal;
and if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is smaller than the second PSNR threshold, judging that the to-be-detected substation equipment image has corresponding defects.
Preferably, the classification method of the pre-trained convolutional classification network comprises the following steps:
inputting the training set into a convolutional classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
and training the convolutional classification network according to the normal substation equipment image and the training defective substation equipment image based on the supervised learning method to obtain the pre-trained convolutional classification network.
Preferably, a defect detection result is obtained according to a similarity classification result and a convolution classification result, and the method specifically comprises the following steps:
and taking the intersection of the similarity classification result and the convolution classification result as a defect detection result.
The invention also provides a small sample substation equipment visual defect detection system based on the improved AnoGAN, which comprises the following steps:
the acquisition module is used for acquiring the training substation equipment image; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
the AnoGAN countermeasure generation network training module is used for training the AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
the preprocessing module is used for inputting the images of the defective substation equipment into the AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
the similarity classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained similarity classification network for classification and outputting a similarity classification result;
the convolution classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained convolution classification network for classification and outputting convolution classification results;
and the defect detection module is used for obtaining a defect detection result according to the similarity classification result and the convolution classification result.
In order to better illustrate the technical effects of the present invention, the present invention provides the following specific embodiments to illustrate the above technical flow:
in embodiment 1, a small sample substation equipment visual defect detection system based on improved AnoGAN, wherein a monitoring model can be divided into an AnoGAN generation model for normal equipment pictures, a convolutional neural network for classifying generation results, and a network for calculating similarity between generated pictures and original pictures, and the detection flow of the whole system is as follows:
a small sample transformer substation equipment visual defect detection method based on improved AnoGAN comprises the following steps:
the first step: dividing the data set into two types of normal equipment and defective equipment (namely normal substation equipment images and defective substation equipment images) according to actual requirements, wherein if defects of various equipment are required to be identified, the data set is required to be further divided according to the equipment types;
and a second step of: training an AnoGAN challenge-generating network using normal substation equipment images (device pictures without defects), wherein the network is composed of a generator and a discriminator, the generator being responsible for generating artificial pictures (target substation equipment images) from the data set, the discriminator being responsible for authenticating the target substation equipment images with the normal substation equipment images;
and a third step of: after the countermeasure generation network reaches balance, calculating peak signal-to-noise ratio (PSNR) of the target substation equipment image and the normal substation equipment image, if the average PSNR value is more than 28, considering that the network can be used, otherwise, repeating the second training step until the PSNR value meets the requirement;
fourth step: fixing the parameters of the countermeasure generation network as a sub-model of a defect detection algorithm;
fifth step: constructing a defect detection model, wherein an input picture (the defect substation equipment image obtained in the first step) is preprocessed by using a countermeasure generation network with fixed parameters in the fourth step, and the result is respectively input into a classification model based on a convolution network and a classification model based on similarity calculation;
sixth step: setting a similarity calculation network parameter, wherein the similarity calculation network calculates PSNR of the target substation equipment image (the preprocessed image obtained in the fifth step) and the normal substation equipment image, and at the moment, a PSNR threshold value is required to be set, and a picture smaller than the threshold value is judged to be a defective equipment and is judged to be a normal equipment. Selecting a test data set of more than 1 ten thousand sheets, respectively calculating and generating a picture PSNR value, setting a PSNR threshold value to limit the judgment accuracy to 99%, and retraining a second step to generate a network if the threshold value is less than 15;
seventh step: training a convolutional classification network, wherein the classification network is trained based on a supervised learning method by using all the equipment data (normal substation equipment images and defective substation equipment images) selected in the first step, and the network can perform defect judgment on the feature pictures output by the generation model;
eighth step: integrating a detection network, wherein the whole network firstly inputs pictures into an countermeasure generation network, the output of the generation network is respectively input into a convolution classification network and a similarity classification network, and the classification result output by the network takes an intersection as a final network judgment result;
ninth step: the system should extract the detected normal substation equipment image and the detected defective substation equipment image every three months and use the extracted normal substation equipment image and the defective substation equipment image to update the detection model so as to ensure the judging effect of the model on the new scene picture.
The embodiment consists of a countermeasure generation network aiming at a normal picture, a convolution classification network aiming at output of the countermeasure network and a PSNR similarity calculation network;
the embodiment provides a network training process for detecting equipment defects aiming at an countermeasure generation network, and provides key indexes of each step in the process to determine whether the current network is available;
the present embodiment proposes a network structure of a discriminator of a neural network for defect detection, wherein a down_block block of the discriminator includes 3X3 convolutions and normal layers, and the picture size becomes 1/2 every time the down_block passes, the input picture size becomes 8 from 128 after the discriminator passes four times the down_block. An additional 1x1 convolution layer is introduced into the traditional full-connection discrimination network to increase the network generalization capability and avoid the great increase of the node number caused by the increase of the post-classification quantity, and the great compression of the neural network can also improve the extraction capability of the network to the characteristic data; as shown in fig. 2.
In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and the division of modules, or units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units, modules, or components may be combined or integrated into another apparatus, or some features may be omitted, or not performed.
The units may or may not be physically separate, and the components shown as units may be one physical unit or a plurality of physical units, may be located in one place, or may be distributed in a plurality of different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.
In particular, according to embodiments of the present disclosure, the processes described above with reference to flowcharts may be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable medium, the computer program comprising program code for performing the method shown in the flowcharts. In such embodiments, the computer program may be downloaded and installed from a network via a communication portion, and/or installed from a removable medium. The above-described functions defined in the method of the present invention are performed when the computer program is executed by a Central Processing Unit (CPU). The computer readable medium of the present invention may be a computer readable signal medium or a computer readable storage medium, or any combination of the two. The computer readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above.
The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
The foregoing is merely illustrative of specific embodiments of the present invention, and the scope of the present invention is not limited thereto, but any changes or substitutions within the technical scope of the present invention should be covered by the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (10)
1. The visual defect detection method for the small sample transformer substation equipment based on the improved AnoGAN is characterized by comprising the following steps of:
acquiring an image of training substation equipment; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
training an AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
inputting the images of the defective substation equipment into an AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
inputting the preprocessed image of the defective substation equipment into a pre-trained similarity classification network for classification, and outputting a similarity classification result;
inputting the preprocessed defective substation equipment image into a pre-trained convolution classification network for classification, and outputting a convolution classification result;
and obtaining a defect detection result according to the similarity classification result and the convolution classification result.
2. The improved AnoGAN-based small sample substation equipment visual defect detection method according to claim 1, wherein the AnoGAN challenge-generation network comprises a generator and a discriminator, the AnoGAN challenge-generation network is trained according to normal substation equipment images, and the AnoGAN challenge-generation network is obtained after training is completed, specifically comprising the following steps:
inputting the normal substation equipment image into a generator of an AnoGAN countermeasure generation network to obtain a target substation equipment image;
inputting the normal substation equipment image and the target substation equipment image into a discriminator of an AnoGAN countermeasure generation network to obtain a discrimination result;
and optimizing a generator according to the distinguishing result to obtain the AnoGAN countermeasure generation network after training is completed.
3. The improved AnoGAN-based small sample substation equipment visual defect detection method according to claim 2, wherein the normal substation equipment image and the target substation equipment image are input into a discriminator of an AnoGAN countermeasure generation network to obtain a discrimination result, specifically comprising the following steps:
calculating peak signal-to-noise ratio PSNR of each normal substation equipment image and each target substation equipment image;
and calculating to obtain a PSNR average value according to the peak signal-to-noise ratio PSNR of all the normal substation equipment images and the target substation equipment images, and taking the PSNR average value as a distinguishing result.
4. The improved AnoGAN-based small sample substation equipment visual defect detection method according to claim 3, wherein the generator is optimized according to the discrimination result to obtain the AnoGAN countermeasure generation network after training is completed, and specifically comprises the following steps:
optimizing network parameters of the generator according to the PSNR average value to obtain an optimized generator;
judging whether the PSNR average value is larger than a first PSNR threshold value or not;
if the PSNR average value is less than or equal to the first PSNR threshold value, retraining the AnoGAN countermeasure generation network;
and ending the optimization until the PSNR average value is larger than the first PSNR threshold value, and obtaining the AnoGAN countermeasure generation network after the training is completed according to the optimized generator.
5. The improved AnoGAN-based small sample substation equipment vision defect detection method according to claim 1, wherein the training method of the pre-trained similarity classification network is as follows:
inputting the training set into a similarity classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image for training and the defective substation equipment image for training;
setting a second PSNR threshold according to the peak signal-to-noise ratio PSNR of the training normal substation equipment image and the training defective substation equipment image;
and obtaining the similarity classification network after pre-training according to the second PSNR threshold value.
6. The improved AnoGAN-based small sample substation equipment visual defect detection method of claim 5, wherein:
and the second PSNR threshold value is smaller than or equal to 99% of peak signal-to-noise ratio PSNR of the training normal substation equipment images and the corresponding training defect substation equipment images in the training set.
7. The improved AnoGAN-based small sample substation equipment visual defect detection method according to claim 6, wherein the preprocessed defective substation equipment image is input into a pre-trained similarity classification network for classification, and a similarity classification result is output, specifically comprising the following steps:
inputting the normal substation equipment image and the preprocessed defective substation equipment image into a pre-trained similarity classification network;
calculating peak signal-to-noise ratio PSNR of the normal substation equipment image and the preprocessed defective substation equipment image;
judging whether the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than a second PSNR threshold value or not;
if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is larger than or equal to a second PSNR threshold value, judging that the to-be-detected substation equipment image is normal;
and if the peak signal-to-noise ratio PSNR of the to-be-detected substation equipment image and the preprocessed defective substation equipment image is smaller than the second PSNR threshold, judging that the to-be-detected substation equipment image has corresponding defects.
8. The improved AnoGAN-based small sample substation equipment visual defect detection method of claim 1, wherein the pre-trained post-convolution classification network classification method is as follows:
inputting the training set into a convolutional classification network; the training set comprises a normal substation equipment image for training and a defective substation equipment image for training;
and training the convolutional classification network according to the normal substation equipment image and the training defective substation equipment image based on the supervised learning method to obtain the pre-trained convolutional classification network.
9. The improved AnoGAN-based visual defect detection method for small-sample substation equipment according to claim 1, wherein the defect detection result is obtained according to a similarity classification result and a convolution classification result, and specifically comprises the following steps:
and taking the intersection of the similarity classification result and the convolution classification result as a defect detection result.
10. A small sample substation equipment visual defect detection system based on improved AnoGAN for implementing the small sample substation equipment visual defect detection method based on improved AnoGAN according to any of claims 1-9, comprising:
the acquisition module is used for acquiring the training substation equipment image; the training substation equipment image comprises a normal substation equipment image and a defect substation equipment image;
the AnoGAN countermeasure generation network training module is used for training the AnoGAN countermeasure generation network according to the normal substation equipment image to obtain the AnoGAN countermeasure generation network after training is completed;
the preprocessing module is used for inputting the images of the defective substation equipment into the AnoGAN countermeasure generation network after training is completed, and outputting the preprocessed images of the defective substation equipment;
the similarity classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained similarity classification network for classification and outputting a similarity classification result;
the convolution classification network classification module is used for inputting the preprocessed defective substation equipment images into a pre-trained convolution classification network for classification and outputting convolution classification results;
and the defect detection module is used for obtaining a defect detection result according to the similarity classification result and the convolution classification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310826080.5A CN116862867A (en) | 2023-07-06 | 2023-07-06 | Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310826080.5A CN116862867A (en) | 2023-07-06 | 2023-07-06 | Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116862867A true CN116862867A (en) | 2023-10-10 |
Family
ID=88233549
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310826080.5A Pending CN116862867A (en) | 2023-07-06 | 2023-07-06 | Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116862867A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117173385A (en) * | 2023-10-24 | 2023-12-05 | 四川思极科技有限公司 | Detection method, device, medium and equipment of transformer substation |
-
2023
- 2023-07-06 CN CN202310826080.5A patent/CN116862867A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117173385A (en) * | 2023-10-24 | 2023-12-05 | 四川思极科技有限公司 | Detection method, device, medium and equipment of transformer substation |
| CN117173385B (en) * | 2023-10-24 | 2024-01-26 | 四川思极科技有限公司 | Detection method, device, medium and equipment of transformer substation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977921B (en) | Method for detecting hidden danger of power transmission line | |
| KR101822829B1 (en) | Method of identifying partial discharge and noise of switchgear using machine learning | |
| CN105844238A (en) | Method and system for discriminating videos | |
| CN109741328A (en) | A kind of automobile apparent mass detection method based on production confrontation network | |
| CN111862065B (en) | Power transmission line diagnosis method and system based on multitask deep convolutional neural network | |
| CN116862867A (en) | Small sample transformer substation equipment visual defect detection method and system based on improved AnoGAN | |
| CN112541393A (en) | Transformer substation personnel detection method and device based on deep learning | |
| CN109993122A (en) | A kind of pedestrian based on depth convolutional neural networks multiplies staircase anomaly detection method | |
| CN111612718A (en) | Human face image restoration method introducing attention mechanism | |
| CN113642474A (en) | Hazardous area personnel monitoring method based on YOLOV5 | |
| CN106960435A (en) | A kind of double compression automatic testing methods of jpeg image | |
| CN108986091A (en) | Casting defect image detecting method based on depth Hash network | |
| CN116486240A (en) | Application of image recognition algorithm in intelligent inspection method of unmanned aerial vehicle of power transmission line | |
| CN114708518A (en) | Bolt defect detection method based on semi-supervised learning and priori knowledge embedding strategy | |
| CN114881665A (en) | Method and system for identifying electricity stealing suspected user based on target identification algorithm | |
| Zheng et al. | Anomaly detection of metro station tracks based on sequential updatable anomaly detection framework | |
| CN113469938A (en) | Pipe gallery video analysis method and system based on embedded front-end processing server | |
| Daogang et al. | Anomaly identification of critical power plant facilities based on YOLOX-CBAM | |
| CN117350964A (en) | Cross-modal multi-level feature fusion-based power equipment detection method | |
| CN116630989A (en) | Visual fault detection method and system for intelligent ammeter, electronic equipment and storage medium | |
| Wu et al. | Memory-augment convolutional Autoencoder for unsupervised pavement crack classification | |
| CN115690584A (en) | SSD (solid State drive) -based improved power distribution room foreign matter detection method | |
| CN115526407A (en) | Power grid big data safety detection early warning method and system based on causal machine learning | |
| CN115273009A (en) | Road crack detection method and system based on deep learning | |
| CN115049875A (en) | Detection method for wearing insulating gloves in transformer substation based on deep learning |
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 |