Distribution network intelligent electricity room defect identification and early warning method based on deep learning

Abstract

The invention relates to a distribution network intelligent electricity room defect identification and early warning method and device based on deep learning, computer equipment, a computer readable storage medium and a computer program product, and can be used in the technical field of artificial intelligence. The method comprises the following steps: acquiring a historical image of the intelligent electric room; according to the defect type of the historical image, performing feature extraction processing on the historical image to obtain defect feature information; performing defect labeling processing on the historical image according to the defect type to obtain a target historical image; performing iterative training on a defect identification model to be trained corresponding to the defect type by using the target historical image and the defect feature information to obtain a target defect identification model; inputting the current image into a target defect identification model for defect identification processing to obtain a defect identification result; and under the condition that the defect identification result shows that the intelligent electricity room has the defect, performing early warning processing on the intelligent electricity room. By adopting the method, the intelligent electric room defect identification and early warning efficiency can be improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment, computer-readable storage medium, and computer program product for defect identification and early warning of intelligent power rooms in distribution networks based on deep learning. Background Art

[0002] With the development of smart grid technology, smart power rooms, as a crucial component of distribution networks, have seen widespread adoption of intelligent monitoring technologies. During the operation and maintenance of smart power rooms, timely detection and resolution of equipment defects are crucial for ensuring power supply security. Therefore, efficient fault identification and early warning in smart power rooms has become a key research topic.

[0003] Traditional technology usually uses manual inspections to identify and warn of defects in smart power rooms. However, this method requires a lot of manual processing time, resulting in low efficiency in identifying and warning defects in smart power rooms. Summary of the Invention

[0004] Based on this, it is necessary to provide a distribution network smart power room defect identification and early warning method, device, computer equipment, computer-readable storage medium and computer program product based on deep learning that can improve the efficiency of smart power room defect identification and early warning in response to the above technical problems.

[0005] In a first aspect, the present application provides a method for identifying and warning defects in a distribution network smart power room based on deep learning. The method comprises:

[0006] Obtain historical images of the smart power room;

[0007] performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0008] According to the defect type, the historical image is subjected to defect labeling processing to obtain a target historical image of the smart power room;

[0009] Using the target historical image and the defect feature information, iteratively training the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room;

[0010] Inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image;

[0011] When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

[0012] In one embodiment, the defect types include animal invasion type, water immersion type, smoke type and flame type;

[0013] The step of performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image includes:

[0014] identifying, from the historical images, a first historical image corresponding to the animal invasion type, a second historical image corresponding to the flooding type, a third historical image corresponding to the smoke type, and a fourth historical image corresponding to the flame type;

[0015] performing feature extraction processing on the first historical image to obtain animal invasion defect feature information of the first historical image;

[0016] performing feature extraction processing on the second historical image to obtain water immersion defect feature information of the second historical image;

[0017] performing feature extraction processing on the third historical image to obtain smoke defect feature information of the third historical image;

[0018] performing feature extraction processing on the fourth historical image to obtain flame defect feature information of the fourth historical image;

[0019] The animal invasion defect feature information, the water immersion defect feature information, the smoke defect feature information, and the flame defect feature information are all used as the defect feature information of the historical image.

[0020] In one embodiment, the performing defect annotation processing on the historical image according to the defect type to obtain the target historical image of the smart power room includes:

[0021] Performing animal invasion defect labeling processing on the first historical image to obtain a first target historical image of the smart power room;

[0022] Performing water immersion defect labeling processing on the second historical image to obtain a second target historical image of the smart power room;

[0023] Performing smoke defect labeling processing on the third historical image to obtain a third target historical image of the smart power room;

[0024] performing flame defect annotation processing on the fourth historical image to obtain a fourth target historical image of the smart power room;

[0025] The first target historical image, the second target historical image, the third target historical image and the fourth target historical image are all used as target historical images of the smart power room.

[0026] In one embodiment, the defect recognition models to be trained corresponding to the defect types include a first defect recognition model to be trained corresponding to the animal intrusion type, a second defect recognition model to be trained corresponding to the water immersion type, a third defect recognition model to be trained corresponding to the smoke type, and a fourth defect recognition model to be trained corresponding to the flame type;

[0027] The method of iteratively training the defect recognition model to be trained corresponding to the defect type by using the target historical image and the defect feature information to obtain the target defect recognition model of the smart power room includes:

[0028] Iteratively training the first defect recognition model to be trained using the first target historical image and the animal intrusion defect feature information to obtain an animal intrusion defect recognition model for the smart power room;

[0029] Iteratively training the second defect recognition model to be trained using the second target historical image and the water flooding defect feature information to obtain a water flooding defect recognition model for the smart power room;

[0030] Iteratively training the third defect recognition model to be trained using the third target historical image and the smoke defect feature information to obtain a smoke defect recognition model for the smart power room;

[0031] Iteratively training the fourth defect recognition model to be trained using the fourth target historical image and the flame defect feature information to obtain a flame defect recognition model for the smart power room;

[0032] The animal intrusion defect recognition model, the water immersion defect recognition model, the smoke defect recognition model and the flame defect recognition model are combined to obtain a target defect recognition model of the smart power room.

[0033] In one embodiment, inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image includes:

[0034] Performing feature extraction processing on the current image to obtain current feature information of the current image;

[0035] The current image and the current feature information are input into the target defect recognition model for defect recognition processing to obtain the defect recognition result.

[0036] In one embodiment, when the defect identification result indicates that the smart power room has a defect, performing early warning processing on the smart power room includes:

[0037] If the defect identification result indicates that the smart power room has a defect, determining a current defect type of the smart power room according to the defect identification result;

[0038] Perform early warning processing on the smart power room according to the current defect type.

[0039] In a second aspect, the present application also provides a distribution network intelligent power room defect identification and early warning device based on deep learning. The device includes:

[0040] Image acquisition module, used to obtain historical images of the smart power room;

[0041] A feature extraction module is used to perform feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0042] a defect annotation module, configured to perform defect annotation processing on the historical image according to the defect type to obtain a target historical image of the smart power room;

[0043] A model training module is used to iteratively train the defect recognition model to be trained corresponding to the defect type using the target historical image and the defect feature information to obtain a target defect recognition model of the smart power room;

[0044] An image input module, configured to input a current image of the smart power room into the target defect recognition model for defect recognition processing, thereby obtaining a defect recognition result of the current image;

[0045] The power room early warning module is used to perform early warning processing on the smart power room when the defect identification result indicates that the smart power room has a defect.

[0046] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are performed:

[0047] Obtain historical images of the smart power room;

[0048] performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0049] According to the defect type, the historical image is subjected to defect labeling processing to obtain a target historical image of the smart power room;

[0050] Using the target historical image and the defect feature information, iteratively training the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room;

[0051] Inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image;

[0052] When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

[0053] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the following steps:

[0054] Obtain historical images of the smart power room;

[0055] performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0056] According to the defect type, the historical image is subjected to defect labeling processing to obtain a target historical image of the smart power room;

[0057] Using the target historical image and the defect feature information, iteratively training the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room;

[0058] Inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image;

[0059] When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

[0060] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the following steps:

[0061] Obtain historical images of the smart power room;

[0062] performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0063] According to the defect type, the historical image is subjected to defect labeling processing to obtain a target historical image of the smart power room;

[0064] Using the target historical image and the defect feature information, iteratively training the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room;

[0065] Inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image;

[0066] When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

[0067] The above-mentioned distribution network smart power room defect identification and early warning method, device, computer equipment, computer-readable storage medium and computer program product based on deep learning obtains historical images of the smart power room; performs feature extraction processing on the historical images according to the defect type of the historical images to obtain defect feature information of the historical images; performs defect labeling processing on the historical images according to the defect type to obtain the target historical images of the smart power room; uses the target historical images and the defect feature information to iteratively train the defect recognition model to be trained corresponding to the defect type to obtain the target defect recognition model of the smart power room; inputs the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain the defect recognition result of the current image; when the defect recognition result indicates that there is a defect in the smart power room, performs early warning processing on the smart power room. This solution obtains historical images of the smart power room and performs feature extraction and annotation processing, providing a reliable data basis for subsequent model training, which is conducive to improving the accuracy of defect identification; by using the annotated historical images and extracted feature information to iteratively train the defect recognition model, the model can accurately identify various defect characteristics, which is conducive to improving the intelligent level of defect identification; by inputting the current image into the trained model for real-time recognition and timely warning when defects are found, it is conducive to the rapid discovery and timely handling of defects in the smart power room, thereby improving the efficiency and accuracy of defect identification and warning in the smart power room. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 1. A flow chart of a method for identifying and warning defects in a distribution network smart power room based on deep learning in one embodiment;

[0070] Figure 2 A flowchart of the steps of determining defect characteristic information in one embodiment;

[0071] Figure 3 Schematic diagram of a flow chart of a method for identifying and warning defects in a distribution network smart power room based on deep learning in another embodiment;

[0072] Figure 4 This is a structural block diagram of a distribution network intelligent power room defect identification and early warning device based on deep learning in one embodiment;

[0073] Figure 5 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0075] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0076] In an exemplary embodiment, Figure 1 As shown, a method for defect identification and early warning of intelligent power rooms in distribution networks based on deep learning is provided. This embodiment uses the method applied to a terminal as an example for illustration; it is understandable that the method can also be applied to a server, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. Among them, the terminal can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, etc.; the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. In this embodiment, the method includes the following steps:

[0077] Step S101: Acquire historical images of the smart power room.

[0078] Step S102 : performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image.

[0079] Step S103 : performing defect annotation processing on the historical image according to the defect type to obtain a target historical image of the smart power room.

[0080] Step S104 , using the target historical image and defect feature information, iteratively train the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room.

[0081] Step S105 : inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image.

[0082] Step S106: When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

[0083] Among them, the smart power room can be a smart power room in the power distribution system. For example, the smart power room can be a power equipment place including various types of Internet of Things sensor terminals and other equipment.

[0084] The historical image may be image data of a smart power room. For example, the historical image may be image data of a smart power room obtained through multiple channels such as smart terminal data, business management data, and externally introduced data.

[0085] The defect type may be various hidden danger types that may occur in the smart power room, for example, the defect type may be small animal intrusion, water immersion, smoke, flame, and the like.

[0086] Among them, feature extraction processing can be the process of performing feature analysis and extraction on image data. For example, feature extraction processing can be extracting the shape features, texture features, and color features of small animals, or extracting the reflective features, color features, and texture features of water immersion, or extracting the transparency features, texture features, and color features of smoke, or extracting the color features, shape features, and dynamic features of flames, etc.

[0087] The defect feature information may be feature data obtained through feature extraction processing. For example, the defect feature information may be shape and texture features extracted using a HOG (Histogram of Oriented Gradients) algorithm, or color features obtained through color histogram statistics.

[0088] Among them, defect labeling processing can be the process of marking and classifying image data. For example, defect labeling processing can be marking the species information of small animals, or marking the location and range of flooded areas, or marking the location and concentration of smoke areas, or marking the location, range, fire size and other information of flames and smoke.

[0089] The target historical image may be image data that has been annotated. For example, the target historical image may be image data that has been annotated with specific defect types and location information.

[0090] The defect recognition model can be a deep learning model used to identify defects in smart power rooms. For example, the defect recognition model can be a target detection model such as the YOLO (You Only Look Once, a deep learning model for real-time target detection) series, Faster-RCNN (Fast Region Convolutional Neural Network), or an image segmentation model such as Mask R-CNN (Mask Region Convolutional Neural Network).

[0091] The current image may be an image captured in real time by the smart power room. For example, the current image may be a monitoring image of the smart power room captured in real time by monitoring equipment in the smart power room.

[0092] Among them, the defect recognition result can be the analysis result of the model on the current image. For example, the defect recognition result can be the judgment result of whether there are defects such as small animals, water immersion, smoke, flames, etc. in the current image.

[0093] The early warning process may be an alarm measure triggered when a defect is detected, for example, the early warning process may be triggering an alarm signal and corresponding protective measures.

[0094] Optionally, the terminal obtains historical images of the smart power room from three types of data: smart terminal data, business management data, and externally imported data. Smart terminal data consists of data signals uploaded by various IoT sensor terminals and other devices. The terminal cleans and enhances the historical images of the smart power room, and also performs cropping and scaling on them. Based on the defect type in the historical images, the terminal extracts defect feature information such as shape, texture, and color. Shape and texture features are extracted using the oriented gradient histogram algorithm, while color features are obtained through color histogram statistics. The terminal annotates the historical images of the smart power room according to defect type, obtaining target historical images annotated with defect type and location information. The terminal inputs the target historical images and defect feature information into a target detection model, such as the YOLO series and Faster-RCNN, for training to obtain a target defect recognition model. The terminal then inputs the current image of the smart power room into the target defect recognition model for recognition processing, obtaining a defect recognition result for the current image and triggering an alert when a defect is detected.

[0095] For example, the terminal captures historical images of smart power rooms that include images of animal intrusion (such as small animals), flooding, smoke, and flames. For images of small animal intrusion, the terminal enhances the image by adjusting contrast and brightness. For images of flooding, the terminal uses a Gaussian filter for pre-smoothing and converts the image to grayscale. For smoke images, the terminal performs grayscale processing and adjusts the dynamic range to enhance the contrast between the smoke area and the background. For flame images, the terminal ensures color accuracy and performs deblurring. The terminal extracts features for each type of image: for small animal images, shape features such as outline and body proportions, texture features such as fur texture, and color distribution are extracted; for flooding images, reflective features, color and texture features of water accumulation areas are extracted; for smoke images, transparency features, flow texture features, and color features are extracted; and for flame images, color features, shape features, and dynamic features are extracted. The terminal annotates these images: the type of small animal, the location and extent of flooding areas, the location and concentration of smoke areas, and the location, extent, and size of the flames and smoke. The terminal uses this labeled data to train a deep learning model to achieve real-time monitoring of the current image of the smart power room. When any type of defect is identified, it immediately triggers corresponding alarms and protective measures.

[0096] In the above-mentioned distribution network smart power room defect identification and early warning method based on deep learning, historical images of the smart power room are obtained; according to the defect type of the historical image, feature extraction processing is performed on the historical image to obtain defect feature information of the historical image; according to the defect type, defect labeling processing is performed on the historical image to obtain the target historical image of the smart power room; using the target historical image and defect feature information, the defect recognition model to be trained corresponding to the defect type is iteratively trained to obtain the target defect recognition model of the smart power room; the current image of the smart power room is input into the target defect recognition model for defect recognition processing to obtain the defect recognition result of the current image; when the defect recognition result indicates that there is a defect in the smart power room, early warning processing is performed on the smart power room. This solution obtains historical images of the smart power room and performs feature extraction and annotation processing, providing a reliable data basis for subsequent model training, which is conducive to improving the accuracy of defect identification; by using the annotated historical images and extracted feature information to iteratively train the defect recognition model, the model can accurately identify various defect characteristics, which is conducive to improving the intelligent level of defect identification; by inputting the current image into the trained model for real-time recognition and timely warning when defects are found, it is conducive to the rapid discovery and timely handling of defects in the smart power room, thereby improving the efficiency and accuracy of defect identification and warning in the smart power room.

[0097] In an exemplary embodiment, the defect types include animal intrusion type, flooding type, smoke type, and flame type; Figure 2, according to the defect type of the historical image, feature extraction processing is performed on the historical image to obtain the defect feature information of the historical image, which specifically includes the following contents:

[0098] Step S201 , identifying, from the historical images, a first historical image corresponding to an animal invasion type, a second historical image corresponding to a flood type, a third historical image corresponding to a smoke type, and a fourth historical image corresponding to a flame type;

[0099] Step S202, performing feature extraction processing on the first historical image to obtain animal invasion defect feature information of the first historical image;

[0100] Step S203, performing feature extraction processing on the second historical image to obtain water immersion defect feature information of the second historical image;

[0101] Step S204: performing feature extraction processing on the third historical image to obtain smoke defect feature information of the third historical image;

[0102] Step S205, performing feature extraction processing on the fourth historical image to obtain flame defect feature information of the fourth historical image;

[0103] In step S206 , the animal invasion defect feature information, the water immersion defect feature information, the smoke defect feature information, and the flame defect feature information are all used as defect feature information of the historical image.

[0104] The animal intrusion type may be an animal (small animal) intrusion situation in the smart power room. For example, the animal intrusion type may be a situation where various animals (small animals) enter the smart power room.

[0105] The flooding type may be water accumulation in the smart power room, for example, the flooding type may be water accumulation of varying degrees on the ground.

[0106] The smoke type may be a situation in which smoke appears in the smart power room, for example, the smoke type may be a situation in which smoke of various concentrations appears.

[0107] The flame type may be a fire situation occurring in the smart power room, for example, the flame type may be a fire situation of various sizes.

[0108] The first historical image may be historical image data containing animal invasion situations. For example, the first historical image may be a historical image recording an animal invasion of a smart power room.

[0109] The second historical image may be historical image data containing flooding conditions. For example, the second historical image may be a historical image recording ground dryness and different levels of water accumulation.

[0110] The third historical image may be historical image data containing smoke conditions. For example, the third historical image may be a historical image recording a smoke-free state and smoke states of various concentrations.

[0111] The fourth historical image may be historical image data containing fire conditions. For example, the fourth historical image may be a historical image recording a normal state, a smoking state, and fire conditions of various sizes.

[0112] Among them, the animal invasion defect feature information can be feature data extracted from the first historical image. For example, the animal invasion defect feature information can be the animal's shape features (such as outline, body proportions), texture features (such as hair texture, feather texture) and color features (color distribution of various animals).

[0113] The water immersion defect feature information may be feature data extracted from the second historical image. For example, the water immersion defect feature information may be reflective features, color features, and texture features of the waterlogged area.

[0114] The smoke defect feature information may be feature data extracted from the third historical image. For example, the smoke defect feature information may be transparency features, texture features (such as blurred, flowing textures), and color features of the smoke.

[0115] Among them, the flame defect feature information can be feature data extracted from the fourth historical image. For example, the flame defect feature information can be the color features (such as high temperature colors such as red, orange, yellow), shape features (such as area, circumference, height, etc.) and dynamic features of the flame.

[0116] Optionally, the terminal first classifies the historical images of the smart power room according to animal intrusion type, flooding type, smoke type, and flame type, obtaining first, second, third, and fourth historical images. The terminal then performs image enhancement processing on the first historical image, adjusting contrast and brightness. It then uses a histogram of oriented gradients algorithm to extract the shape, texture, and color features of small animals in the first historical image, obtaining animal intrusion defect feature information. The terminal then performs Gaussian filtering pre-smoothing on the second historical image and converts it into a grayscale image. It then extracts the reflective, color, and texture features of the waterlogged area, obtaining water flooding defect feature information. The terminal then performs grayscale processing on the third historical image and adjusts the dynamic range to enhance the contrast between the smoke area and the background. It then extracts the transparency, texture, and color features of the smoke, obtaining smoke defect feature information. The terminal then performs color accuracy assurance and deblurring on the fourth historical image, extracting the color, shape, and dynamic features of the flame, obtaining flame defect feature information. The terminal then integrates the animal intrusion defect feature information, the flooding defect feature information, the smoke defect feature information, and the flame defect feature information into defect feature information for the historical images.

[0117] The technical solution provided in this embodiment is conducive to targeted feature extraction of different types of defects by clearly dividing the defect types of the smart power room into animal invasion type, water immersion type, smoke type and flame type, and identifying the corresponding first historical image, second historical image, third historical image and fourth historical image from the historical images respectively; by performing feature extraction processing on each type of historical image separately, the corresponding defect feature information is obtained, and these feature information are integrated, which is conducive to obtaining complete defect feature information, thereby improving the accuracy and comprehensiveness of defect identification in the smart power room.

[0118] In an exemplary embodiment, defect labeling is performed on historical images according to defect types to obtain target historical images of the smart power room, specifically including the following: animal invasion defect labeling is performed on the first historical image to obtain the first target historical image of the smart power room; water immersion defect labeling is performed on the second historical image to obtain the second target historical image of the smart power room; smoke defect labeling is performed on the third historical image to obtain the third target historical image of the smart power room; flame defect labeling is performed on the fourth historical image to obtain the fourth target historical image of the smart power room; the first target historical image, the second target historical image, the third target historical image and the fourth target historical image are all used as target historical images of the smart power room.

[0119] Among them, the animal invasion defect labeling processing can be a processing process of marking and classifying images containing animal (small animal) invasion situations. For example, the animal invasion defect labeling processing can be a processing process of labeling whether there are animals (small animals) in the image and the type of animal information.

[0120] Among them, the water immersion defect labeling process can be a processing process of marking and classifying images containing water accumulation conditions. For example, the water immersion defect labeling process can be a processing process of labeling images of ground dryness and various water accumulation levels, and labeling the location and range of the water accumulation area.

[0121] Among them, the smoke defect labeling process can be a process of marking and classifying images containing smoke conditions. For example, the smoke defect labeling process can be a process of labeling images of smoke-free states and smoke states of various concentrations, and marking the location and concentration of the smoke area.

[0122] Among them, flame defect labeling processing can be a process of marking and classifying images containing fire conditions. For example, flame defect labeling processing can be a process of labeling images of normal state, smoking state and fire state (various fire sizes), and labeling information such as the location, range, and fire size of the flame and smoke.

[0123] The first target historical image may be a historical image that has been annotated with animal invasion defects. For example, the first target historical image may be a historical image that has been annotated with animal species information.

[0124] The second target historical image may be a historical image that has been annotated with water immersion defects. For example, the second target historical image may be a historical image in which the location and range of the waterlogged area have been annotated.

[0125] The third target historical image may be a historical image that has been annotated with smoke defects. For example, the third target historical image may be a historical image in which the location and concentration of the smoke area have been annotated.

[0126] The fourth target historical image may be a historical image that has been processed by flame defect annotation. For example, the fourth target historical image may be a historical image that has been annotated with information such as the location, range, and size of the flame and smoke.

[0127] Optionally, the terminal performs animal intrusion defect labeling on the first historical image of the smart power room, marking whether animals exist in the first historical image and information about the species of animals, to obtain a first target historical image of the smart power room; the terminal performs water flooding defect labeling on the second historical image of the smart power room, marking the dryness of the ground and various water accumulation levels in the second historical image, and marking the location and range of the water accumulation area, to obtain a second target historical image of the smart power room; the terminal performs smoke defect labeling on the third historical image of the smart power room, marking the smoke-free state and various smoke concentrations in the third historical image, and marking the location and concentration of the smoke area, to obtain a third target historical image of the smart power room; the terminal performs flame defect labeling on the fourth historical image of the smart power room, marking the normal state, smoking state and fire state in the fourth historical image, and marking information about the location, range and size of the flame and smoke, to obtain a fourth target historical image of the smart power room; the terminal integrates the first target historical image, the second target historical image, the third target historical image and the fourth target historical image of the smart power room into the target historical image of the smart power room.

[0128] The technical solution provided in this embodiment labels the historical images of the smart power room according to the four defect types of animal intrusion, water immersion, smoke and flame, thereby obtaining corresponding target historical images, which is conducive to establishing a dedicated labeling data set for each defect type; by independently labeling each type of historical image and integrating all labeled target historical images, it is conducive to obtaining complete and clearly classified defect labeling data, thereby improving the accuracy and comprehensiveness of defect identification in the smart power room.

[0129] In an exemplary embodiment, the defect recognition models to be trained corresponding to the defect types include a first defect recognition model to be trained corresponding to the animal intrusion type, a second defect recognition model to be trained corresponding to the water immersion type, a third defect recognition model to be trained corresponding to the smoke type, and a fourth defect recognition model to be trained corresponding to the flame type; using the target historical image and defect feature information, the defect recognition models to be trained corresponding to the defect types are iteratively trained to obtain the target defect recognition model of the smart power room, which specifically includes the following contents: using the first target historical image and animal invasion defect feature information, the first defect recognition model to be trained is iteratively trained to obtain the animal invasion defect recognition model of the smart power room. Recognition model; using the second target historical image and water immersion defect feature information, the second defect recognition model to be trained is iteratively trained to obtain the water immersion defect recognition model of the smart power room; using the third target historical image and smoke defect feature information, the third defect recognition model to be trained is iteratively trained to obtain the smoke defect recognition model of the smart power room; using the fourth target historical image and flame defect feature information, the fourth defect recognition model to be trained is iteratively trained to obtain the flame defect recognition model of the smart power room; the animal intrusion defect recognition model, the water immersion defect recognition model, the smoke defect recognition model and the flame defect recognition model are combined to obtain the target defect recognition model of the smart power room.

[0130] Among them, the first defect recognition model to be trained can be an untrained deep learning model for identifying animal invasion situations. For example, the first defect recognition model to be trained can be a target detection model such as the YOLO series or Faster-RCNN.

[0131] The second defect recognition model to be trained may be an untrained deep learning model for identifying water immersion conditions. For example, the second defect recognition model to be trained may be an image segmentation model such as Mask R-CNN.

[0132] The third defect recognition model to be trained may be an untrained deep learning model for identifying smoke conditions. For example, the third defect recognition model to be trained may be a convolutional neural network (CNN) or a model based on a combination of manual features and a machine learning classifier.

[0133] The fourth defect recognition model to be trained may be an untrained deep learning model for identifying flame conditions. For example, the fourth defect recognition model to be trained may be a multi-target detection model that combines target detection and classification functions.

[0134] Among them, the animal intrusion defect recognition model can be a deep learning model trained to identify animal intrusion situations. For example, the animal intrusion defect recognition model can be a target detection model trained to identify animal characteristics and location information.

[0135] The water immersion defect recognition model may be a deep learning model that has been trained to identify water immersion conditions. For example, the water immersion defect recognition model may be an image segmentation model that has been trained to accurately identify waterlogged areas.

[0136] The smoke defect recognition model may be a deep learning model that has been trained to identify smoke conditions. For example, the smoke defect recognition model may be a convolutional neural network model that has been trained to accurately identify smoke.

[0137] Among them, the flame defect recognition model can be a deep learning model trained to identify flame conditions. For example, the flame defect recognition model can be a multi-target detection and classification model trained to accurately identify fire conditions.

[0138] Among them, the target defect recognition model can be a comprehensive model that combines all trained defect recognition models. For example, the target defect recognition model can be a comprehensive model that can identify multiple defect types by integrating the animal intrusion defect recognition model, the water immersion defect recognition model, the smoke defect recognition model and the flame defect recognition model.

[0139] Optionally, the terminal first prepares defect recognition models to be trained, including a target detection model such as the YOLO series or Faster-RCNN for identifying animal intrusion as the first defect recognition model to be trained, an image segmentation model such as Mask R-CNN for identifying water flooding as the second defect recognition model to be trained, a convolutional neural network model for identifying smoke as the third defect recognition model to be trained, and a multi-target detection and classification model for identifying flames as the fourth defect recognition model to be trained. The terminal iteratively trains the first defect recognition model to be trained using the first target historical images and animal intrusion defect feature information to obtain an animal intrusion defect recognition model; iteratively trains the second defect recognition model to be trained using the second target historical images and water flooding defect feature information to obtain a water flooding defect recognition model; iteratively trains the third defect recognition model to be trained using the third target historical images and smoke defect feature information to obtain a smoke defect recognition model; and iteratively trains the fourth defect recognition model to be trained using the fourth target historical images and flame defect feature information to obtain a flame defect recognition model. The terminal combines these four trained defect recognition models to obtain the target defect recognition model for the smart power room.

[0140] The technical solution provided in this embodiment is beneficial to improving the recognition accuracy of each defect type by setting up special models to be trained for the four defect types of animal intrusion, water immersion, smoke and flame, and using the corresponding type of target historical images and defect feature information to perform targeted iterative training on each model; by combining the four trained special defect recognition models to obtain the target defect recognition model of the smart power room, it is beneficial to achieve unified recognition of multiple defect types, thereby improving the comprehensiveness and accuracy of defect recognition in the smart power room.

[0141] In an exemplary embodiment, the current image of the smart power room is input into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image, which specifically includes the following contents: feature extraction processing is performed on the current image to obtain current feature information of the current image; the current image and the current feature information are input into the target defect recognition model for defect recognition processing to obtain a defect recognition result.

[0142] Among them, the current feature information can be various types of defect feature data extracted from the current image. For example, the current feature information can be the shape features of animal invasion (such as the outline and body proportions of small animals), the reflective features and color features of water immersion, the transparency features and texture features of smoke, the color features of flames (such as high-temperature colors such as red, orange, and yellow) and shape features (such as area, circumference, height, etc.) extracted from the current image, and other information.

[0143] Among them, the defect recognition result can be the defect type and status judgment output by the target defect recognition model after analyzing the current image. For example, the defect recognition result can be the specific recognition results such as whether there is animal invasion and its type, whether there is water flooding and its scope, whether there is smoke and its concentration, whether there is flame and its size in the current image.

[0144] Optionally, the terminal performs feature extraction processing on the current image of the smart power room, extracting multi-dimensional feature information such as shape features, texture features, color features, and reflective features from the current image to obtain current feature information of the current image of the smart power room. The terminal inputs the current image and the current feature information into a target defect recognition model for the smart power room. The target defect recognition model analyzes the current image to determine whether defects such as animal intrusion, water immersion, smoke, or flames are present in the current image, thereby obtaining a defect recognition result for the current image of the smart power room.

[0145] The technical solution provided in this embodiment obtains current feature information by performing feature extraction processing on the current image of the smart power room, and inputs the current image and current feature information into the target defect recognition model. By using the original image data and the extracted feature information as dual inputs, it is beneficial to improve the efficiency and accuracy of defect recognition, thereby facilitating the accurate recognition and judgment of defects in the smart power room.

[0146] In an exemplary embodiment, when the defect identification result indicates that there is a defect in the smart power room, early warning processing is performed on the smart power room, which specifically includes the following contents: when the defect identification result indicates that there is a defect in the smart power room, the current defect type of the smart power room is determined according to the defect identification result; and according to the current defect type, early warning processing is performed on the smart power room.

[0147] The current defect type may be a specific defect type currently existing in the smart power room identified by the target defect recognition model. For example, the current defect type may be an animal intrusion type (such as a small animal entering the power room), a water immersion type (such as water on the ground), a smoke type (such as equipment smoking), or a flame type (such as equipment catching fire).

[0148] Among them, early warning processing can be alarms and protective measures taken according to the type of defects detected in the smart power room. For example, early warning processing can be issuing an expulsion alarm when an animal intrusion is detected, starting the drainage system when flooding is detected, turning on the ventilation equipment when smoke is detected, triggering the fire protection system when a flame is detected, and other corresponding alarms and protective measures.

[0149] Optionally, after receiving the defect identification result for the smart power room, the terminal first determines whether the defect identification result indicates a defect in the smart power room. If so, the terminal determines, based on the characteristic information contained in the defect identification result, whether the current defect type in the smart power room is animal intrusion, flooding, smoke, or flame. Based on the determined current defect type, the terminal selects a corresponding early warning solution and performs early warning processing for the smart power room, thereby achieving a timely response to the smart power room defect.

[0150] The technical solution provided in this embodiment is conducive to the accurate identification of defects in the smart power room by determining the specific type of the current defect based on the defect identification result when determining that there is a defect in the smart power room, and performing corresponding early warning processing based on the determined defect type; by establishing a corresponding relationship between defect identification and early warning processing, it is conducive to the realization of differentiated early warnings for different types of defects, thereby improving the accuracy and timeliness of early warning processing in the smart power room.

[0151] The following application example illustrates the distribution network smart power room defect identification and early warning method based on deep learning provided by this application. This application example uses the method applied to the terminal as an example.

[0152] As a key component of the power system, distribution room equipment plays a vital role in ensuring a stable and reliable power supply. The safe operation of this equipment is directly related to the stability of the entire power grid and the normal functioning of society. However, potential defects in line equipment often lead to power system failures and power outages, with significant impacts on economic activities and residents' daily lives.

[0153] In current operations and maintenance practices, the inspection and maintenance of distribution room equipment primarily relies on traditional manual inspections. While this method can ensure basic equipment operation to a certain extent, its inherent limitations are becoming increasingly apparent. Manual inspections are not only inefficient and unable to meet the maintenance needs of large-scale power grids, but are also costly, placing a significant financial burden on operators. More importantly, manual inspections often fail to fully cover all equipment, making it even more difficult to effectively identify potential defects within the equipment, increasing the likelihood of failures.

[0154] Main technical issues:

[0155] Traditional distribution network inspections rely primarily on manual labor, resulting in significant issues such as low efficiency, low accuracy, and limited coverage. Furthermore, due to the limitations of manual inspections, some hidden internal equipment defects are often difficult to detect in a timely manner, increasing the risk of equipment failure.

[0156] Minor technical issues:

[0157] (1) Limitations of infrared thermal imaging technology: To make up for the shortcomings of manual inspections, infrared thermal imaging technology is widely used in fault detection of equipment in power distribution rooms. This technology can detect overheating faults in equipment in a timely manner by detecting the temperature distribution on the surface of the equipment. However, infrared thermal imaging technology also faces some limitations, such as high requirements for ambient temperature and limited ability to detect small-scale faults. These limitations restrict its detection effect in certain application scenarios.

[0158] (2) Disadvantages of ultrasonic testing technology: Ultrasonic testing technology is another commonly used method for detecting equipment defects. This technology uses the propagation characteristics of ultrasonic waves in different media to detect defects within equipment. Although ultrasonic testing technology has the advantages of high detection accuracy and no damage to equipment, it is complex to operate and requires professional personnel. In addition, its detection range is limited, making it difficult to meet the maintenance needs of large-scale power grids.

[0159] refer to Figure 3,The main processes of the application example are as follows: data ,collection and processing, data feature extraction, data labeling and ,division, model training, real-time detection and response.

[0160] (1) Data collection and processing:

[0161] Through multi-channel data acquisition and power sample sharing and exchange mechanisms, raw data samples are acquired from three types of data: smart terminal data, business management data, and externally introduced data. Smart terminal data consists of data signals uploaded by various IoT sensor terminals and other devices. The generated data is directly connected to the IoT platform through an intelligent gateway or platform. After identification and labeling, the data is transmitted to the artificial intelligence platform according to unified standards and stored uniformly in the data center. Business management data is generated in real time by various professional management systems. After identification and labeling, the data is transmitted to the artificial intelligence platform according to unified standards and finally aggregated to the data center for storage. Externally introduced data consists of data shared by internal units and introduced by external units. Through the regular exchange of artificial intelligence samples and model sharing, as well as the introduction of external data through the energy industry data space, the sample database is continuously expanded and enriched, and uniformly aggregated to the data center for storage.

[0162] The images include small animals, water immersion, smoke, and flame recognition in smart power rooms. The image data of smart power rooms is cleaned and enhanced, and the images are cropped and scaled. Targeted image processing is performed for each defect content:

[0163] Small animal image processing: Perform image enhancement processing, such as adjusting contrast and brightness, to make the features of small animals more obvious.

[0164] Water immersion image processing: Gaussian filtering and other methods are used to smooth the collected images and reduce noise, especially noise caused by factors such as ground reflection. Color images are converted to grayscale images to simplify subsequent processing.

[0165] Smoke image processing: Perform image preprocessing and grayscale processing to convert color images into grayscale images, simplifying subsequent processing. Based on the image brightness distribution, adjust the image dynamic range and enhance the contrast between the smoke area and the background.

[0166] Flame image processing: Ensures the color accuracy of the image, as flame color is one of the important features for fire identification; uses a deblurring algorithm to improve image clarity for better identification of flame and smoke features.

[0167] (2) Data feature extraction:

[0168] Identify defect types (hazard types) in smart power room image data and extract features based on defect types:

[0169] Small Animal Feature Extraction: Extracts visual features of small animals, including shape (such as the animal's outline and body proportions), texture (such as fur and feather texture), and color (different small animals have different color distributions). Shape and texture features are extracted using the directional gradient histogram algorithm, while color features are obtained by performing color histogram statistics on the image.

[0170] Water immersion feature extraction: Extracting reflective features. Waterlogged areas typically have strong reflective properties in images. This is done by analyzing the brightness and reflective patterns of pixels in the image. Color and texture features are also extracted. Waterlogged areas can darken the ground and create a smoother texture. This method extracts the differences in color and texture between the ground with and without water.

[0171] Smoke feature extraction: Smoke has a certain degree of transparency, causing the pixel values in the smoke area of the image to vary within a certain range. Transparency features are extracted by analyzing the distribution and variation of pixel values. Texture features are also extracted. Smoke appears to have a unique, blurry, and flowing texture in the image. Texture analysis methods such as wavelet transform are used to extract the texture features of smoke. Color features are also extracted. In some cases, smoke has a specific color (such as black or gray smoke from a fire). Color features can be extracted to aid in judgment.

[0172] Flame feature extraction: Flames typically appear as high-temperature colors such as red, orange, and yellow. Flame color features are extracted by analyzing the pixel value distribution of specific color channels in the image, such as the RGB (red, green, and blue) or HSV (hue, saturation, and value) color spaces. Flame shape and dynamic features are also extracted. Flames have unique irregular shapes and are constantly changing and moving. Techniques such as optical flow are used to extract the motion characteristics of flames, while also analyzing their shape characteristics (such as area, perimeter, and height). Smoke features are extracted as auxiliary features. Fires are often accompanied by smoke, and smoke-related features (such as transparency, texture, and color) are extracted to assist in fire assessment.

[0173] (3) Data annotation division:

[0174] According to the defect type of the smart power room, the target defect type is processed and marked:

[0175] Small animal image annotation: Annotate the image to see if there are small animals and their species.

[0176] Waterlogging image annotation: Collect and annotate images showing dry ground conditions and varying degrees of waterlogging, and mark the location and extent of waterlogged areas.

[0177] Smoke image annotation: Collect and annotate images containing smoke-free states and smoke states of varying concentrations, marking the location and concentration of smoke areas. Annotation can be assisted by manual judgment or other sensor data.

[0178] Flame image annotation: Collect and annotate images of normal, smoking, and fire states (of varying fire sizes), and annotate information such as the location, range, and fire size estimation of flames and smoke.

[0179] (4) Model training:

[0180] Based on the characteristics of each defect type, a deep learning model is used for training:

[0181] Small animal model training: Use deep learning models, such as the YOLO series and Faster-RCNN object detection models. Extracted features are fed into these models for training, allowing them to learn the characteristics and location information of small animals. Traditional machine learning methods use feature matching algorithms, such as template matching, or feature vector classification algorithms (such as the K-nearest neighbor algorithm), to determine the presence of small animals based on feature similarity.

[0182] Waterlogging model training: Use an image segmentation model, such as Mask R-CNN, to train the model using a labeled image dataset, enabling it to accurately identify waterlogged areas.

[0183] Smoke model training: Using a CNN (convolutional neural network) or a method based on combining handcrafted features with a machine learning classifier, the labeled image dataset is used to train the model, enabling it to accurately identify smoke.

[0184] Fire model training: This uses a multi-target detection and classification model, combining target detection (for detecting flames and smoke) and classification (for determining fire size). This model is trained on a dataset of labeled images, enabling it to accurately identify fire conditions.

[0185] (5) Real-time detection and response:

[0186] In actual operation, newly acquired images undergo preprocessing and feature extraction before being fed into a trained model for detection. If the model output indicates a defect in the smart power room, appropriate alarms and protective measures are triggered.

[0187] The technical solution provided by this application example enables defect identification in smart power rooms, resolving the issues of low efficiency, low accuracy, and limited inspection coverage associated with traditional inspections. The method presented in this application example enables real-time monitoring and early warning of various defects within the power room, improving inspection efficiency and defect identification accuracy, ensuring timely detection and resolution of defects and minimizing false alarms or missed reports due to human error. Furthermore, this application example enables all-weather, all-encompassing monitoring, ensuring comprehensive inspection coverage and providing a more reliable guarantee for the safe and stable operation of the power system.

[0188] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0189] Based on the same inventive concept, the embodiments of the present application also provide a deep learning-based distribution network intelligent power room defect identification and early warning device for implementing the deep learning-based distribution network intelligent power room defect identification and early warning method involved above. The implementation solution provided by this device is similar to the implementation solution described in the above method. Therefore, the specific limitations of one or more embodiments of the deep learning-based distribution network intelligent power room defect identification and early warning device provided below can be referred to the limitations of the deep learning-based distribution network intelligent power room defect identification and early warning method above, and will not be repeated here.

[0190] In an exemplary embodiment, Figure 4 As shown, a distribution network intelligent power room defect identification and early warning device based on deep learning is provided. The distribution network intelligent power room defect identification and early warning device based on deep learning 400 may include:

[0191] Image acquisition module 401, used to acquire historical images of the smart power room;

[0192] The feature extraction module 402 is used to perform feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image;

[0193] The defect annotation module 403 is used to perform defect annotation processing on the historical image according to the defect type to obtain the target historical image of the smart power room;

[0194] The model training module 404 is used to iteratively train the defect recognition model to be trained corresponding to the defect type using the target historical image and defect feature information to obtain the target defect recognition model of the smart power room;

[0195] The image input module 405 is used to input the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain the defect recognition result of the current image;

[0196] The power room early warning module 406 is used to perform early warning processing on the smart power room when the defect identification result indicates that the smart power room has defects.

[0197] In an exemplary embodiment, defect types include animal invasion type, water immersion type, smoke type and flame type; the feature extraction module 402 is further used to identify, from historical images, a first historical image corresponding to the animal invasion type, a second historical image corresponding to the water immersion type, a third historical image corresponding to the smoke type and a fourth historical image corresponding to the flame type; perform feature extraction processing on the first historical image to obtain animal invasion defect feature information of the first historical image; perform feature extraction processing on the second historical image to obtain water immersion defect feature information of the second historical image; perform feature extraction processing on the third historical image to obtain smoke defect feature information of the third historical image; perform feature extraction processing on the fourth historical image to obtain flame defect feature information of the fourth historical image; and use the animal invasion defect feature information, water immersion defect feature information, smoke defect feature information and flame defect feature information as defect feature information of the historical image.

[0198] In an exemplary embodiment, the defect labeling module 403 is also used to perform animal invasion defect labeling on the first historical image to obtain a first target historical image of the smart power room; perform water immersion defect labeling on the second historical image to obtain a second target historical image of the smart power room; perform smoke defect labeling on the third historical image to obtain a third target historical image of the smart power room; perform flame defect labeling on the fourth historical image to obtain a fourth target historical image of the smart power room; and use the first target historical image, the second target historical image, the third target historical image and the fourth target historical image as target historical images of the smart power room.

[0199] In an exemplary embodiment, the defect recognition models to be trained corresponding to the defect types include a first defect recognition model to be trained corresponding to the animal intrusion type, a second defect recognition model to be trained corresponding to the water immersion type, a third defect recognition model to be trained corresponding to the smoke type, and a fourth defect recognition model to be trained corresponding to the flame type; the model training module 404 is further used to iteratively train the first defect recognition model to be trained using the first target historical image and the animal intrusion defect feature information to obtain the animal intrusion defect recognition model of the smart power room; to iteratively train the second defect recognition model to be trained using the second target historical image and the water immersion defect feature information to obtain the water immersion defect recognition model of the smart power room; to iteratively train the third defect recognition model to be trained using the third target historical image and the smoke defect feature information to obtain the smoke defect recognition model of the smart power room; to iteratively train the fourth defect recognition model to be trained using the fourth target historical image and the flame defect feature information to obtain the flame defect recognition model of the smart power room; and to combine the animal intrusion defect recognition model, the water immersion defect recognition model, the smoke defect recognition model, and the flame defect recognition model to obtain the target defect recognition model of the smart power room.

[0200] In an exemplary embodiment, the image input module 405 is also used to perform feature extraction processing on the current image to obtain current feature information of the current image; input the current image and current feature information into the target defect recognition model for defect recognition processing to obtain a defect recognition result.

[0201] In an exemplary embodiment, the power room early warning module 406 is also used to determine the current defect type of the smart power room according to the defect identification result when the defect identification result indicates that there is a defect in the smart power room; and perform early warning processing on the smart power room according to the current defect type.

[0202] Each module in the aforementioned deep learning-based distribution network intelligent power room defect identification and early warning device can be implemented in whole or in part through software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in hardware form, or stored in a computer device memory in software form, so that the processor can call and execute the corresponding operations of each module.

[0203] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 5As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, while the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be achieved via Wi-Fi, mobile cellular networks, NFC (near-field communication), or other technologies. When executed by the processor, the computer program implements a deep learning-based method for defect identification and early warning in a distribution network smart room. The display unit of the computer device is used to produce visual images and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0204] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0205] In an exemplary embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0206] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0207] In an exemplary embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0208] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0209] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0210] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A distribution network intelligent power room defect identification and early warning method based on deep learning, characterized in that: The method comprises: Obtain historical images of the smart power room; performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image; According to the defect type, the historical image is subjected to defect labeling processing to obtain a target historical image of the smart power room; Using the target historical image and the defect feature information, iteratively training the defect recognition model to be trained corresponding to the defect type to obtain a target defect recognition model of the smart power room; Inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image; When the defect identification result indicates that the smart power room has a defect, an early warning process is performed on the smart power room.

2. The method according to claim 1, characterized in that The defect types include animal invasion type, water immersion type, smoke type and flame type; The step of performing feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image includes: identifying, from the historical images, a first historical image corresponding to the animal invasion type, a second historical image corresponding to the flooding type, a third historical image corresponding to the smoke type, and a fourth historical image corresponding to the flame type; performing feature extraction processing on the first historical image to obtain animal invasion defect feature information of the first historical image; performing feature extraction processing on the second historical image to obtain water immersion defect feature information of the second historical image; performing feature extraction processing on the third historical image to obtain smoke defect feature information of the third historical image; performing feature extraction processing on the fourth historical image to obtain flame defect feature information of the fourth historical image; The animal invasion defect feature information, the water immersion defect feature information, the smoke defect feature information, and the flame defect feature information are all used as the defect feature information of the historical image.

3. The method according to claim 2, characterized in that The step of performing defect labeling on the historical image according to the defect type to obtain a target historical image of the smart power room includes: Performing animal invasion defect labeling processing on the first historical image to obtain a first target historical image of the smart power room; Performing water immersion defect labeling processing on the second historical image to obtain a second target historical image of the smart power room; Performing smoke defect labeling processing on the third historical image to obtain a third target historical image of the smart power room; performing flame defect annotation processing on the fourth historical image to obtain a fourth target historical image of the smart power room; The first target historical image, the second target historical image, the third target historical image and the fourth target historical image are all used as target historical images of the smart power room.

4. The method according to claim 3, characterized in that The defect recognition models to be trained corresponding to the defect types include a first defect recognition model to be trained corresponding to the animal intrusion type, a second defect recognition model to be trained corresponding to the water immersion type, a third defect recognition model to be trained corresponding to the smoke type, and a fourth defect recognition model to be trained corresponding to the flame type; The method of iteratively training the defect recognition model to be trained corresponding to the defect type by using the target historical image and the defect feature information to obtain the target defect recognition model of the smart power room includes: Iteratively training the first defect recognition model to be trained using the first target historical image and the animal intrusion defect feature information to obtain an animal intrusion defect recognition model for the smart power room; Iteratively training the second defect recognition model to be trained using the second target historical image and the water flooding defect feature information to obtain a water flooding defect recognition model for the smart power room; Iteratively training the third defect recognition model to be trained using the third target historical image and the smoke defect feature information to obtain a smoke defect recognition model for the smart power room; Iteratively training the fourth defect recognition model to be trained using the fourth target historical image and the flame defect feature information to obtain a flame defect recognition model for the smart power room; The animal intrusion defect recognition model, the water immersion defect recognition model, the smoke defect recognition model and the flame defect recognition model are combined to obtain a target defect recognition model of the smart power room.

5. The method according to claim 1, wherein The step of inputting the current image of the smart power room into the target defect recognition model for defect recognition processing to obtain a defect recognition result of the current image includes: Performing feature extraction processing on the current image to obtain current feature information of the current image; The current image and the current feature information are input into the target defect recognition model for defect recognition processing to obtain the defect recognition result.

6. The method according to any one of claims 1 to 5, characterized in that When the defect identification result indicates that the smart power room has a defect, performing early warning processing on the smart power room includes: If the defect identification result indicates that the smart power room has a defect, determining a current defect type of the smart power room according to the defect identification result; Perform early warning processing on the smart power room according to the current defect type.

7. A distribution network intelligent power room defect identification and early warning device based on deep learning, characterized in that: The device comprises: Image acquisition module, used to obtain historical images of the smart power room; A feature extraction module is used to perform feature extraction processing on the historical image according to the defect type of the historical image to obtain defect feature information of the historical image; a defect annotation module, configured to perform defect annotation processing on the historical image according to the defect type to obtain a target historical image of the smart power room; A model training module is used to iteratively train the defect recognition model to be trained corresponding to the defect type using the target historical image and the defect feature information to obtain a target defect recognition model of the smart power room; An image input module, configured to input a current image of the smart power room into the target defect recognition model for defect recognition processing, thereby obtaining a defect recognition result of the current image; The power room early warning module is used to perform early warning processing on the smart power room when the defect identification result indicates that the smart power room has a defect.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images