CN118097373A - Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel - Google Patents
Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel Download PDFInfo
- Publication number
- CN118097373A CN118097373A CN202410457913.XA CN202410457913A CN118097373A CN 118097373 A CN118097373 A CN 118097373A CN 202410457913 A CN202410457913 A CN 202410457913A CN 118097373 A CN118097373 A CN 118097373A
- Authority
- CN
- China
- Prior art keywords
- transmission channel
- power transmission
- hidden danger
- image feature
- mae
- 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
- 230000005540 biological transmission Effects 0.000 title claims abstract description 88
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000001514 detection method Methods 0.000 claims abstract description 74
- 238000012549 training Methods 0.000 claims abstract description 45
- 230000004927 fusion Effects 0.000 claims abstract description 35
- 230000008569 process Effects 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims description 19
- 230000000007 visual effect Effects 0.000 claims description 11
- 108091026890 Coding region Proteins 0.000 claims description 6
- 238000004590 computer program Methods 0.000 claims description 6
- 230000006872 improvement Effects 0.000 claims description 6
- 230000008014 freezing Effects 0.000 claims description 4
- 238000007710 freezing Methods 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 208000037004 Myoclonic-astatic epilepsy Diseases 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000000874 microwave-assisted extraction Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012876 topography Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Classifications
-
- 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
- Image Analysis (AREA)
Abstract
The invention relates to an unsupervised method, a system and a storage medium for detecting hidden danger of a power transmission channel. The invention creates an image feature encoder and decoder; and constructing an MAE and SimCLR fusion training framework, wherein SimCLR branch results from different deformations of the same image are subjected to feature fusion with coding results of an MAE image feature coder, then the contrast loss is calculated with SimCLR branches from other images, the contrast loss is subjected to weighted fusion with MAE decoding loss, the image feature coder and an improved CASCADE RCNN detection part are combined to create a power transmission channel hidden danger detection large model, power transmission channel scene data training with hidden danger labels is used on the basis of the power transmission channel hidden danger detection large model created by the image feature coder, image feature coder parameters are frozen in the training process, and parameters of a detection part of CASCADE RCNN are improved to train and detect the power transmission channel hidden danger.
Description
Technical Field
The invention relates to the technical field of computer vision for detecting a power transmission channel, in particular to an unsupervised power transmission channel hidden danger detection method, an unsupervised power transmission channel hidden danger detection system and a storage medium.
Background
In recent years, the field of hidden danger detection of a power transmission channel based on deep learning has been remarkably developed. Target detection techniques, such as the two-stage R-CNN series and the single-stage detector YOLO series, are widely used to improve the accuracy and speed of target detection. In order to solve the problem of scarcity of marked data, transfer learning becomes a common strategy, and the performance of the model is improved by pre-training on a large-scale data set and then fine-tuning on transmission channel data.
Despite significant advances made in power transmission channel hidden danger detection models based on supervised training, there are still some potential drawbacks and challenges. Supervised learning models typically require large amounts of labeled data for efficient training. However, in the field of power transmission channel hidden trouble detection, it is very expensive and time-consuming to acquire a sufficient number of high-quality marks. In the scene of hidden danger detection, the hidden danger of different types may have a class imbalance problem, that is, the number of samples of some hidden danger types is far smaller than that of other types, which may cause the model to be too biased to the classes with more number, and the hidden danger detection performance of few classes is poor. If the model is only exposed to data for a particular environment or scene while training, it may perform poorly in an unseen environment.
By means of massive unlabeled power transmission channel data, training of a model by using an unsupervised method becomes possible, and the model is not dependent on specific label types any more, so that the perception capability of potential hidden danger features is improved, a powerful image encoder can be obtained, and the potential hidden danger detection of the power transmission channel is carried out by matching with a detector. The existing unsupervised method trains the encoder to use one of MAE (Masked Autoencoders) or SimCLR (a new framework of visual characterization versus learning), and in fact, the combination of the two is more beneficial to the learning, similar to the learning by human being by simultaneously reading and writing, so that the model learns deeper and richer features.
Disclosure of Invention
In order to solve the technical problems or at least partially solve the technical problems, the invention provides an unsupervised method, an unsupervised system and a storage medium for detecting hidden danger of a power transmission channel.
In a first aspect, the present invention provides an unsupervised method for detecting hidden danger of a power transmission channel, including:
Creating an image feature encoder that supports simultaneous use by both MAE and SimCLR unsupervised learning methods and shares weights; creating a decoder for restoring the features of the image feature encoder in the MAE process to the original input image; constructing an MAE and SimCLR fusion training framework to train the image feature encoder and decoder, wherein SimCLR branch results from different deformations of the same image are subjected to feature fusion with the encoding results of the MAE image feature encoder, then contrast loss is calculated with SimCLR branches from other images, and the contrast loss and MAE decoding loss are subjected to weighted fusion to train the image feature encoder and decoder;
And combining the image feature encoder with the improved CASCADE RCNN detection part to create a power transmission channel hidden danger detection large model, training the power transmission channel scene data with hidden danger labels based on the power transmission channel hidden danger detection large model created by the image feature encoder, freezing parameters of the image feature encoder in the training process, and updating only parameters of the improved CASCADE RCNN detection part, thereby realizing the detection of the power transmission channel hidden danger.
Still further, the image feature encoder is constructed to include: the method comprises the steps of firstly, carrying out local feature extraction by using a convolutional neural network to obtain a CNN feature map, then, carrying out global feature coding by using the convolutional neural network to obtain a transducer coding sequence, and finally, carrying out fusion on the transducer coding sequence and the CNN feature map to obtain accurate image feature coding.
Still further, the decoder is a top-down pyramid network built using CNNs.
Still further, a spatial location awareness module consisting of visual transducers and Resnet is added before each cascade of CASCADE RCNN to form the improvement CASCADE RCNN.
Still further, the training image feature encoder trains the improved CASCADE RCNN detection section to use the power transmission channel scene data of the hidden trouble signature using channel scene data covering various time periods, various weather, and various topography.
Further, the fusion formula of the contrast loss and the MAE decoding loss is as follows:
；
Wherein, 、/>、/>、/>Are all equilibrium coefficients, N is the contrast loss, where/>Represents MAE decoding penalty, using a weighted fusion of L1 penalty and L2 penalty.
In a second aspect, the present invention provides a hidden danger detection system for a power transmission channel, to implement the unsupervised hidden danger detection method for the power transmission channel, including:
The image acquisition module acquires scene data of the power transmission channel;
an encoder-decoder module that creates an image feature encoder that supports simultaneous use by both MAE and SimCLR unsupervised learning methods and that shares weights; creating a decoder for restoring the characteristics of the encoder in the MAE process to the original input image;
a first training module that builds a MAE and SimCLR fusion training framework to train the image feature encoder and decoder, wherein SimCLR branch results from different deformations of the same image are feature fused with the encoding results of the MAE image feature encoder, then a contrast loss is calculated with SimCLR branches from other images, and the contrast loss is weighted fused with MAE decoding loss to train the image feature encoder and decoder;
the model construction module combines the image feature encoder and the improved CASCADE RCNN detection part to create a large power transmission channel hidden danger detection model;
And the second training module uses the power transmission channel scene data with hidden danger labels to train a power transmission channel hidden danger detection large model created based on the image feature encoder, freezes the parameters of the image feature encoder in the training process, and only updates and improves the parameters of the CASCADE RCNN detection part so as to realize the detection of the power transmission channel hidden danger.
In a third aspect, the present invention provides a computer readable storage medium storing a computer program, which when executed by a processor, implements the unsupervised method for detecting hidden danger of a power transmission channel.
Compared with the prior art, the technical scheme provided by the embodiment of the invention has the following advantages:
The invention creates an image feature encoder which supports simultaneous use by two unsupervised learning methods, MAE and SimCLR, and shares weights; creating a decoder for restoring the features of the image feature encoder in the MAE process to the original input image; the MAE and SimCLR fusion training framework is constructed to train the image feature encoder and decoder, wherein SimCLR branch results from different variants of the same image are feature fused with the encoding results of the MAE image feature encoder, then contrast loss is calculated with SimCLR branches from other images, and the contrast loss results are weighted fused with MAE decoding loss to train the image feature encoder and decoder. The ideas of the two unsupervised learning methods MAE and SimCLR are fused to train the image feature encoder, so that the visual features extracted by the image feature encoder are more abundant, the feature encoding capability of the model is improved, and the encoding of hidden danger features in rich scenes is supported.
The invention combines the image feature encoder and the improved CASCADE RCNN detection part to create a power transmission channel hidden danger detection large model, uses power transmission channel scene data training with hidden danger labels to be based on the power transmission channel hidden danger detection large model created by the image feature encoder, freezes the parameters of the image feature encoder in the training process, and only updates the parameters of the detection part of the improved CASCADE RCNN, thereby realizing the detection of the hidden danger of the power transmission channel. The detection part of the improvement CASCADE RCNN is added with a space position sensing module consisting of a visual transducer and Resnet, so that the CASCADE RCNN capability of combining and identifying richer scenes and hidden dangers is improved, and the accurate hidden dangers of the power transmission channel are detected.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
In order to more clearly illustrate the embodiments of the invention or the technical solutions of the prior art, the drawings which are used in the description of the embodiments or the prior art will be briefly described, and it will be obvious to a person skilled in the art that other drawings can be obtained from these drawings without inventive effort.
Fig. 1 is a flowchart of an unsupervised method for detecting hidden danger of a power transmission channel according to an embodiment of the present invention;
FIG. 2 is an exemplary diagram of a framework for fusion training of MAEs and SimCLR provided in an embodiment of the present invention;
FIG. 3 is an exemplary diagram of a large power transmission channel hidden danger detection model provided by an embodiment of the present invention;
Fig. 4 is a schematic diagram of an unsupervised power transmission channel hidden danger detection system according to an embodiment of the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
Example 1
Referring to fig. 1, an embodiment of the present invention provides an unsupervised method for detecting hidden danger of a power transmission channel, including the following steps:
Image feature encoders are created that support simultaneous use by both MAE and SimCLR unsupervised learning methods and share weights. The invention aims to simulate the human learning process, and the ideas of the two unsupervised learning methods MAE (Masked Autoencoders) and SimCLR are fused to train the image feature encoder, so that the extracted visual features are more abundant, the feature encoding capability of a model is improved, and rich and correct visual features are provided for decoding.
In the implementation process, the constructed image feature encoder comprises: convolutional neural networks and a Swin transducer-based transducer encoder. The image feature encoder is used for extracting the features of the scene image of the power transmission channel, and the working principle is as follows: firstly, local feature extraction is carried out by using a convolutional neural network to obtain a CNN feature map, then global feature coding is carried out by using a transducer coder to obtain a transducer coding sequence, and finally, the transducer coding sequence and the CNN feature map are fused to obtain accurate feature coding. An exemplary convolutional neural network adopts Resnet model 101, an exemplary transducer encoder uses a Swin-L transducer model, firstly uses Resnet to extract local features to obtain CNN feature images, then uses the Swin-L transducer model to carry out global feature encoding to obtain a transducer encoding sequence, and finally fuses the transducer encoding sequence and the CNN feature images to obtain accurate feature encoding of images.
A decoder is created for restoring the features of the image feature encoder in the MAE process to the original input image. In the implementation process, in the process MAE (Masked Autoencoders), the image feature encoder extracts the features of the original input image, and the decoder restores the features of the original input image to the original input image. MAE decoding loss is calculated by generating a comparison between the original input image and the true original input image. Calculating MAE decoding penalty uses a weighted fusion of L1 penalty and L2 penalty, i.e.: MAE decoding loss is expressed as. In one example, the decoder is a top-down pyramid network built using CNNs.
Referring to fig. 2, a frame of fusion training is constructed for MAE and SimCLR to train the image feature encoder and decoder, in the frame of fusion training is constructed for MAE and SimCLR, simCLR branch results from different deformations of the same image are feature fused with the encoding results of the MAE image feature encoder, then contrast loss is calculated with SimCLR branches from other images, tensor conversion is performed before contrast loss is calculated with the features of SimCLR branches from other images to conform to the dimension of the fusion feature, and the contrast loss results are fused with the MAE decoding loss, training is performed together, and the final loss is weighted fusion of the contrast loss and the MAE decoding loss, wherein the fusion formula of the contrast loss is as follows:
；
Wherein, 、/>、/>、/>Are all equilibrium coefficients, N is the contrast loss, where/>Representing MAE decoding penalty, using a weighted fusion of L1 penalty and L2 penalty; in one embodiment,/>、/>、/>、/>Respectively set to 0.3,0.7,0.5,0.5.
According to the application, the image feature encoder for extracting the scene image features of the power transmission channel is trained in an unsupervised mode, and the scene image features of the power transmission channel are not required to be marked in the process. Therefore, the scene data of the transmission channel with larger magnitude is supported to train, so that the extracted visual features are richer, the feature coding capacity of the model is improved, and rich and correct visual features are provided for decoding.
In the implementation process, billion-level power transmission channel scene data are constructed to train the image feature encoder according to a training framework. In the implementation process, the channel scene data cover the power transmission channel scenes of various time periods, various weather and various landforms; the time span of the image shot by each device is 3 years and more from the perspective of single shooting device, the time span of the image shot by each device covers 24 hours per day, four seasons of spring, summer, autumn and winter, various weather of yin, sunny, rainy, fog and snowy, and the image shot by the device covers various landforms of plain, sea river, plateau, desert and basin from the perspective of region.
And constructing a large model of hidden danger detection of the transmission channel by using the trained image feature encoder, wherein in the specific implementation process, as shown in fig. 3, the image feature encoder is combined with the improved CASCADE RCNN detection part to create the large model of hidden danger detection of the transmission channel. For the features extracted by the image feature encoder, the detecting part is modified CASCADE RCNN to identify whether hidden trouble exists. In the present application, the improvement CASCADE RCNN detection section adds a spatial location awareness module consisting of visual transducers and Resnet to the front of each cascade of CASCADE RCNN to form the improvement CASCADE RCNN. The characteristics extracted by the image characteristic encoder can reflect the abundant scene information and hidden danger information of the power transmission channel, and provide abundant and correct visual characteristics for decoding. In order to improve CASCADE RCNN combining recognition capability to richer scenes and hidden dangers, a spatial position sensing module is added in front of each cascade head of CASCADE RCNN, different detection parts can pay attention to different information, effective information is utilized to the greatest extent, and therefore detection capability of a model is improved.
And (3) training the power transmission channel scene data with hidden danger labels based on the power transmission channel hidden danger detection large model created by the image feature encoder, freezing the image feature encoder parameters in the training process, and updating only the parameters of the detection part of the improvement CASCADE RCNN. And detecting the hidden danger of the transmission channel through the trained hidden danger detection large model of the transmission channel.
Example 2
Referring to fig. 4, an embodiment of the present invention provides an image enhancement system for a power transmission channel, to implement the method for detecting hidden danger of an unsupervised power transmission channel, including:
The image acquisition module acquires scene data of the power transmission channel; the training image feature encoder is acquired by the image acquisition module using billions of magnitude of transmission channel scene data covering various time periods, various weather and various landforms.
And the marking module marks the scene data of the power transmission channel containing hidden danger to form the scene data of the power transmission channel for training the hidden danger label of the improved CASCADE RCNN detection part.
An encoder-decoder module that creates an image feature encoder that supports simultaneous use by both MAE and SimCLR unsupervised learning methods and that shares weights; creating a decoder for restoring the characteristics of the encoder in the MAE process to the original input image;
A first training module that builds a framework for fusion training of MAE and SimCLR to train the image feature encoder and decoder, wherein SimCLR branch results from different deformations of the same image are feature fused with the encoding results of the MAE image feature encoder, then the contrast loss is calculated with SimCLR branches from other images, and the contrast loss results are fused with MAE decoding loss, training is performed together, and the final loss is a weighted fusion of the contrast loss and MAE decoding loss,
The fusion formula of the contrast loss and the MAE decoding loss is as follows:
；
Wherein, 、/>、/>、/>Are all equilibrium coefficients, N is the contrast loss, where/>Represents MAE decoding penalty, using a weighted fusion of L1 penalty and L2 penalty.
And the model construction module combines the image feature encoder and the improved CASCADE RCNN detection part to create a large power transmission channel hidden danger detection model.
And the second training module uses the power transmission channel scene data with hidden danger labels to train a power transmission channel hidden danger detection large model created based on the image feature encoder, freezes the parameters of the image feature encoder in the training process, and only updates and improves the parameters of the CASCADE RCNN detection part so as to realize the detection of the power transmission channel hidden danger.
Example 3
The embodiment of the invention provides a computer readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the unsupervised hidden danger detection method of the power transmission channel is realized.
The method comprises the following steps:
Creating an image feature encoder that supports simultaneous use by both MAE and SimCLR unsupervised learning methods and shares weights; creating a decoder for restoring the features of the image feature encoder in the MAE process to the original input image; constructing a framework for fusion training of MAE and SimCLR to train the image feature encoder and decoder, wherein MAE encoding results are subjected to feature fusion with SimCLR input of branches which are different in deformation and come from the same image, then calculating contrast loss with SimCLR input of branches which are not come from the same image, and carrying out weighted fusion on the contrast loss results and MAE decoding loss to train the image feature encoder and decoder; the fusion formula of the contrast loss and the MAE decoding loss is as follows:
；
Wherein, 、/>、/>、/>Are all equilibrium coefficients, N is the contrast loss, where/>Represents MAE decoding penalty, using a weighted fusion of L1 penalty and L2 penalty.
And combining the image feature encoder with the improved CASCADE RCNN detection part to create a power transmission channel hidden danger detection large model, training the power transmission channel scene data with hidden danger labels based on the power transmission channel hidden danger detection large model created by the image feature encoder, freezing parameters of the image feature encoder in the training process, and updating only parameters of the detection part of the improved CASCADE RCNN so as to realize detection of the hidden danger of the power transmission channel.
Of course, the computer readable storage medium provided by the embodiment of the present invention stores a computer program not limited to the above-mentioned method operations, but also can execute the related operations in the unsupervised power transmission channel hidden danger detection method provided by any embodiment of the present invention.
In the embodiments provided in the present invention, it should be understood that the disclosed structures and methods may be implemented in other manners. For example, the structural embodiments described above are merely illustrative, and for example, the division of the units is merely a logical function division, and there may be other manners of division in actual implementation, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via interfaces, structures or units, which may be in electrical, mechanical or other forms.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
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.
The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. An unsupervised method for detecting hidden danger of a power transmission channel is characterized by comprising the following steps:
Creating an image feature encoder that supports simultaneous use by both MAE and SimCLR unsupervised learning methods and shares weights; creating a decoder for restoring the features of the image feature encoder in the MAE process to the original input image; constructing an MAE and SimCLR fusion training framework to train the image feature encoder and decoder, wherein SimCLR branch results from different deformations of the same image are subjected to feature fusion with the encoding results of the MAE image feature encoder, then contrast loss is calculated with SimCLR branches from other images, and the contrast loss and MAE decoding loss are subjected to weighted fusion to train the image feature encoder and decoder;
And combining the image feature encoder with the improved CASCADE RCNN detection part to create a power transmission channel hidden danger detection large model, training the power transmission channel scene data with hidden danger labels based on the power transmission channel hidden danger detection large model created by the image feature encoder, freezing the image feature encoder parameters which are used for two unsupervised learning training modes of MAE and SimCLR, and only updating the parameters of the detection part of the improved CASCADE RCNN so as to obtain the power transmission channel hidden danger detection large model, and detecting the power transmission channel hidden danger.
2. The unsupervised power transmission channel hidden danger detection method according to claim 1, wherein the constructed image feature encoder comprises: the method comprises the steps of firstly, carrying out local feature extraction by using a convolutional neural network to obtain a CNN feature map, then, carrying out global feature coding by using the convolutional neural network to obtain a transducer coding sequence, and finally, carrying out fusion on the transducer coding sequence and the CNN feature map to obtain accurate image feature coding.
3. The method for unsupervised power transmission channel hidden danger detection according to claim 1, wherein the decoder is a top-down pyramid network built using CNN.
4. The method of claim 1, wherein the improvement CASCADE RCNN is formed by adding a spatial location awareness module consisting of visual transducers and Resnet before each cascading head of CASCADE RCNN.
5. The unsupervised power transmission channel hidden danger detection method according to claim 1, wherein the training image feature encoder uses channel scene data covering various time periods, various weather and various landforms, and the training the improved CASCADE RCNN detection section uses power transmission channel scene data of hidden danger tags.
6. The method for detecting hidden danger of an unsupervised power transmission channel according to claim 1, wherein the fusion formula of the contrast loss and the MAE decoding loss is as follows:
；
Wherein, 、/>、/>、/>Are all equilibrium coefficients, N is the contrast loss, where/>Represents MAE decoding penalty, using a weighted fusion of L1 penalty and L2 penalty.
7. A hidden danger detection system for a power transmission channel, implementing the unsupervised power transmission channel hidden danger detection method according to any one of claims 1 to 6, comprising:
The image acquisition module acquires scene data of the power transmission channel;
an encoder-decoder module providing an image feature encoder supporting simultaneous use by both MAE and SimCLR unsupervised learning methods and sharing weights; creating a decoder for restoring the characteristics of the encoder in the MAE process to the original input image;
a first training module that builds a MAE and SimCLR fusion training framework to train the image feature encoder and decoder, wherein SimCLR branch results from different deformations of the same image are feature fused with the encoding results of the MAE image feature encoder, then a contrast loss is calculated with SimCLR branches from other images, and the contrast loss is weighted fused with MAE decoding loss to train the image feature encoder and decoder;
the model construction module combines the image feature encoder and the improved CASCADE RCNN detection part to create a large power transmission channel hidden danger detection model;
And the second training module uses the power transmission channel scene data with hidden danger labels to train a power transmission channel hidden danger detection large model created based on the image feature encoder, freezes the parameters of the image feature encoder in the training process, and only updates and improves the parameters of the CASCADE RCNN detection part so as to realize the detection of the power transmission channel hidden danger.
8. A computer readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the unsupervised power transmission channel hidden danger detection method according to any one of claims 1-6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410457913.XA CN118097373A (en) | 2024-04-17 | 2024-04-17 | Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410457913.XA CN118097373A (en) | 2024-04-17 | 2024-04-17 | Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118097373A true CN118097373A (en) | 2024-05-28 |
Family
ID=91142122
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410457913.XA Pending CN118097373A (en) | 2024-04-17 | 2024-04-17 | Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118097373A (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114627360A (en) * | 2020-12-14 | 2022-06-14 | 国电南瑞科技股份有限公司 | Substation equipment defect identification method based on cascade detection model |
| CN116977763A (en) * | 2022-12-28 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Model training method, device, computer readable storage medium and computer equipment |
| WO2024032010A1 (en) * | 2022-08-11 | 2024-02-15 | 重庆邮电大学 | Transfer learning strategy-based real-time few-shot object detection method |
-
2024
- 2024-04-17 CN CN202410457913.XA patent/CN118097373A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114627360A (en) * | 2020-12-14 | 2022-06-14 | 国电南瑞科技股份有限公司 | Substation equipment defect identification method based on cascade detection model |
| WO2024032010A1 (en) * | 2022-08-11 | 2024-02-15 | 重庆邮电大学 | Transfer learning strategy-based real-time few-shot object detection method |
| CN116977763A (en) * | 2022-12-28 | 2023-10-31 | 腾讯科技（深圳）有限公司 | Model training method, device, computer readable storage medium and computer equipment |
Non-Patent Citations (2)
| Title |
|---|
| Y CHEN ET AL.: "Target detection based on improved swin transformer and cascade RCNN", 《3RD INTERNATIONAL CONFERENCE ON ARTIFICIAL INTELLIGENCE, VIRTUAL REALITY AND VISUALIZATION (AIVRV 2023)》, vol. 12923, 8 November 2023 (2023-11-08), XP060193354, DOI: 10.1117/12.3011400 * |
| 刘亚琳: "基于深度学习的车辆目标检测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》, no. 11, 15 November 2022 (2022-11-15) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111563508B (en) | Semantic segmentation method based on spatial information fusion | |
| Zhang et al. | Remote sensing image spatiotemporal fusion using a generative adversarial network | |
| CN111915592B (en) | Remote sensing image cloud detection method based on deep learning | |
| Xue et al. | Remote sensing scene classification based on multi-structure deep features fusion | |
| CN113780149B (en) | Remote sensing image building target efficient extraction method based on attention mechanism | |
| CN110147794A (en) | A kind of unmanned vehicle outdoor scene real time method for segmenting based on deep learning | |
| CN113936339A (en) | Fighting identification method and device based on double-channel cross attention mechanism | |
| CN112446342B (en) | Key frame recognition model training method, recognition method and device | |
| CN109635726B (en) | Landslide identification method based on combination of symmetric deep network and multi-scale pooling | |
| CN115223067B (en) | Point cloud fusion method, device and equipment applied to unmanned aerial vehicle and storage medium | |
| CN116363526B (en) | MROCNet model construction and multisource remote sensing image change detection method and system | |
| CN115761222B (en) | Image segmentation method, remote sensing image segmentation method and device | |
| CN111291705B (en) | Pedestrian re-identification method crossing multiple target domains | |
| CN113486886A (en) | License plate recognition method and device in natural scene | |
| CN104463962B (en) | Three-dimensional scene reconstruction method based on GPS information video | |
| CN116524189A (en) | High-resolution remote sensing image semantic segmentation method based on coding and decoding indexing edge characterization | |
| CN117370498B (en) | Unified modeling method for 3D open vocabulary detection and closed caption generation | |
| CN113901931A (en) | Knowledge distillation model-based behavior recognition method for infrared and visible light videos | |
| CN116912673A (en) | Target detection method based on underwater optical image | |
| CN112330562A (en) | Heterogeneous remote sensing image transformation method and system | |
| CN118097373A (en) | Unsupervised method, unsupervised system and storage medium for detecting hidden danger of power transmission channel | |
| CN113128461B (en) | Pedestrian re-recognition performance improving method based on human body key point mining full-scale features | |
| CN115131206A (en) | Semantic understanding-based satellite video super-resolution reconstruction method and system | |
| CN115331171A (en) | Crowd counting method and system based on depth information and significance information | |
| CN114882449B (en) | Car-Det network model-based vehicle detection method and device |
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 |