CN-122023407-A - Information identification method and information identification system for transformer substation protection device

Abstract

The application discloses an information identification method and an information identification system for a transformer substation protection device, and belongs to the technical field of computer vision. The method comprises the steps of obtaining a short exposure image and a long exposure image of a transformer substation protection device at the same moment, obtaining a basic feature map based on the short exposure image and the long exposure image, carrying out layered feature extraction on the basic feature map according to a global attention network, a local attention network and a super local attention network, generating a normalized attention map, generating a fusion feature map based on the short exposure image, the long exposure image and the normalized attention map, carrying out dynamic weight distribution fusion on the short exposure image and the long exposure image to obtain an enhanced image, carrying out stain area detection and stain repair treatment on the enhanced image, outputting a stain removal image, identifying tripping codes and indicator lamp states from the stain removal image, and outputting information identification results. The embodiment of the application can effectively improve the accuracy of identifying the tripping information of the transformer substation protection device.

Inventors

• CHEN QIONGLIANG
• WU BAOXING
• LI NA
• XU YALE
• ZHENG JUANJUAN
• CHEN YING
• ZHAO ZIYI
• ZHENG WEI
• YE FENG
• XU JIYAO
• LIN GAOXIANG
• YE LIZHAO
• CHEN LI
• CHEN GANG
• PAN SUSU
• Chen Jituo
• WAN XIAO

Assignees

• 国网浙江省电力有限公司温州供电公司

Dates

Publication Date

20260512

Application Date

20260413

Claims (10)

1. 1. An information identification method of a transformer substation protection device is characterized by comprising the following steps: Acquiring a short exposure image and a long exposure image of a transformer substation protection device at the same moment, wherein the exposure parameters of the short exposure image and the long exposure image are different; Obtaining a basic feature map based on the short exposure image and the long exposure image, carrying out layered feature extraction on the basic feature map according to a global attention network, a local attention network and a super local attention network, generating a normalized attention map, and generating a fusion feature map based on the short exposure image, the long exposure image and the normalized attention map; Based on the fusion feature map and the normalized attention map, carrying out dynamic weight distribution fusion on the short exposure image and the long exposure image to obtain an enhanced image; Performing stain area detection and stain repair treatment on the enhanced image, and outputting a stain-removing image; And identifying the tripping code and the state of the indicator lamp from the stain removal image, and outputting an information identification result.
2. 2. The method of claim 1, wherein the performing hierarchical feature extraction on the base feature map according to a global attention network, a local attention network, and a super local attention network, generating a normalized attention map, and generating a fused feature map based on the short exposure image, the long exposure image, and the normalized attention map comprises: generating corresponding global, local and hyperlocal attention profiles via the global, local and hyperlocal attention networks based on the base feature map; Normalizing the global attention map, the local attention map and the super local attention map to obtain a normalized attention map, wherein the normalized attention map comprises a normalized global attention map, a normalized local attention map and a normalized super local attention map.
3. 3. The method of claim 2, wherein performing a stain region detection and stain repair process on the enhanced image to output a stain-removed image comprises: The enhanced image is matched with the fusion feature image pixel by pixel, and whether the pixel point value of the fusion feature image is smaller than a preset threshold value or not is judged to be subjected to primary screening, so that a stain candidate area is obtained; Determining a suspected stain region by calculating the difference between the gray scale and the gray scale average value of pixels in the stain candidate region; Based on whether the suspected stain area is overlapped with the character core area of the normalized local attention seeking to mark or the edge area of the normalized super local attention seeking to mark, storing a non-overlapped part and obtaining a real stain area; And repairing the non-edge stain area in the real stain area by adopting neighborhood average filtering, and simultaneously repairing the stain area near the edge in the real stain area by adopting bilateral filtering, so as to output a stain removing image.
4. 4. The method of claim 2, wherein the identifying trip codes and indicator light status from the stain removal image and outputting information identification results comprises: extracting a connected region of the normalized local attention map from the stain removal image to obtain a character candidate region, and recognizing tripping codes and equipment number characters by adopting a CRNN model after local self-adaptive binarization and median filtering pretreatment of the character candidate region; Extracting a closed communication region of the normalized super-local attention map from the stain removal image, combining region area screening to obtain a real indicator light region, and classifying indicator light states by adopting an SVM model after calculating color features and brightness features of the real indicator light region; And constructing an information identification result based on the tripping code and the indicator light state.
5. 5. The method of claim 4, wherein said constructing an information identification result based on said trip code and said indicator light status comprises: cross-verifying the tripping code and the state of the indicator lamp to obtain a verification result; if the verification result is verification passing, an information identification result is constructed based on the verification result passing verification.
6. 6. The method of claim 1, wherein the deriving a base feature map based on the short exposure image and the long exposure image comprises: And respectively extracting basic features from the short exposure image and the long exposure image through the first 3 layers of ResNet-18 networks, and performing cross-exposure fusion to obtain a basic feature map.
7. 7. An information identification system of a transformer substation protection device, to which the information identification method of the transformer substation protection device according to claims 1 to 6 is applied, characterized in that the system comprises a hardware layer and a software layer; the hardware layer comprises the following modules connected in sequence: The image acquisition module is used for synchronously acquiring a short exposure image and a long exposure image of the transformer substation protection device; The processing module is used for running the software layer to obtain an information identification result; The storage module is used for storing the information identification result; The communication module is used for carrying out data interaction with an external operation and maintenance platform and outputting the information identification result; The software layer comprises the following modules connected in sequence: The multi-exposure acquisition control module is used for generating an acquisition trigger signal for controlling the image acquisition module to acquire a short exposure image and a long exposure image of the transformer substation protection device at the same moment, and the exposure parameters of the short exposure image and the long exposure image are different; The multi-scale attention feature extraction module is used for obtaining a basic feature map based on the short exposure image and the long exposure image, carrying out layered feature extraction on the basic feature map according to a global attention network, a local attention network and a super local attention network, generating a normalized attention map, and generating a fusion feature map based on the short exposure image, the long exposure image and the normalized attention map; the dynamic fusion enhancement module is used for carrying out dynamic weight distribution fusion on the short exposure image and the long exposure image based on the fusion feature map and the normalized attention map to obtain an enhanced image; The stain detection and shielding module is used for carrying out stain region detection and stain repair treatment on the enhanced image and outputting a stain removal image; And the tripping information identification module is used for identifying tripping codes and indicator lamp states from the stain-removing images and outputting information identification results.
8. 8. The system of claim 7, wherein the image acquisition module is configured to synchronously acquire a short exposure image and a long exposure image of the same scene of the substation protection device, the image acquisition module comprising a light supplementing unit for providing illumination for the scene.
9. 9. The system of claim 7, wherein the multi-exposure acquisition control module is integrated with a target detection algorithm, the multi-exposure acquisition control module is configured to generate an acquisition trigger signal based on a processing result of the target detection algorithm, and send the acquisition trigger signal to the image acquisition module to control the image acquisition module to synchronously acquire a short-exposure image and a long-exposure image, and receive the short-exposure image and the long-exposure image returned by the image acquisition module.
10. 10. The system of claim 7, wherein the system comprises a plurality of sensors, A global attention network provided with a 16 x 16 convolution kernel configured to capture cabinet global layout features and output a global attention map; A local attention network provided with an 8 x 8 convolution kernel configured to focus character semantic features and output a local attention map; a hyperlocal attention network provided with a4 x 4 convolution kernel configured to capture character edge gradient features and output a hyperlocal attention map.

Description

Information identification method and information identification system for transformer substation protection device Technical Field The application relates to the technical field of computer vision, in particular to an information identification method and an information identification system for a transformer substation protection device. Background The transformer substation protection device is important equipment for safe operation of the power system, tripping information (such as display screen characters and indicator light states) of the transformer substation protection device directly reflects fault types and fault positions, and is an important basis for operation and maintenance personnel to quickly isolate faults and restore power supply. The prior art generally relies on a fixed camera to collect trip information, and the fixed camera is easily affected by environmental interference, so that the trip information cannot be accurately identified. Disclosure of Invention The embodiment of the application aims to provide an information identification method and an information identification system for a transformer substation protection device, which can effectively improve the accuracy of tripping information identification of the transformer substation protection device. To achieve the above object, a first aspect of an embodiment of the present application provides an information identification method of a substation protection device, including: Acquiring a short exposure image and a long exposure image of a transformer substation protection device at the same moment, wherein the exposure parameters of the short exposure image and the long exposure image are different; Obtaining a basic feature map based on the short exposure image and the long exposure image, carrying out layered feature extraction on the basic feature map according to a global attention network, a local attention network and a super local attention network, generating a normalized attention map, and generating a fusion feature map based on the short exposure image, the long exposure image and the normalized attention map; Based on the fusion feature map and the normalized attention map, carrying out dynamic weight distribution fusion on the short exposure image and the long exposure image to obtain an enhanced image; Performing stain area detection and stain repair treatment on the enhanced image, and outputting a stain-removing image; And identifying the tripping code and the state of the indicator lamp from the stain removal image, and outputting an information identification result. Compared with the prior art, the information identification method for the transformer substation protection device has the advantages that short exposure and long exposure images of the transformer substation protection device are synchronously acquired, high-light area details and dark area information can be reserved respectively, a global-local-super-local three-layer attention network is used for carrying out layered extraction on basic feature images to generate a normalized attention map, global layout and local details can be simultaneously considered, dynamic weight distribution fusion can be realized based on the fusion feature map and the normalized attention map, fusion weights of the long exposure images and the short exposure images can be adaptively adjusted for different areas of the images, the limitation that the existing fixed weight fusion is difficult to adapt to local light differences of the transformer substation is overcome, enhanced images are output, spot detection and repair are carried out in a targeted mode, shielding interference of spots such as screen cabinet dust and oil spots on information areas can be eliminated, trip code and indicator lamp states are finally identified, the obtained information identification result can still keep high accuracy under complex environment interference, trip information identification accuracy of the transformer substation protection device is effectively improved, and further operation response efficiency and safe operation guarantee capacity of a power grid are effectively improved. In some embodiments, the performing hierarchical feature extraction on the base feature map according to a global attention network, a local attention network, and a super local attention network, generating a normalized attention map, and generating a fused feature map based on the short exposure image, the long exposure image, and the normalized attention map, includes: generating corresponding global, local and hyperlocal attention profiles via the global, local and hyperlocal attention networks based on the base feature map; Normalizing the global attention map, the local attention map and the super local attention map to obtain a normalized attention map, wherein the normalized attention map comprises a normalized global attention map, a normalized local attention map and a normalized sup