Power equipment image identification safety early warning method and equipment

Abstract

The invention discloses a power equipment image identification safety early warning method and equipment, and the method comprises the steps: carrying out the target detection of preprocessed image data through an improved YOLO + Transform model, enhancing the distinguishing capability of a target and a background through the introduction of a Transform module, employing a multi-head self-attention mechanism, optimizing the local feature expression in combination with a sliding window attention mechanism, and achieving the safety early warning of the image identification of the power equipment. According to the method, the detection stability in a complex environment is improved, BiFPN feature fusion and DLA deep feature aggregation are adopted to optimize target detection, small target information is effectively transmitted among feature layers of different scales, so that missing detection and false detection are reduced, in addition, K-Means + + is adopted to optimize the Anchor frame size, and a Soft Anchor Assignment strategy is combined, so that target matching is more accurate, and the recall rate of small targets is further improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent monitoring and safety warning of power equipment, and particularly relates to a method and device for image recognition safety warning of power equipment. Background Art

[0002] The safe operation of power equipment is crucial to the stability of the power grid. With the continuous expansion of the power grid scale, the inspection work of transmission lines, substations and distribution equipment has become increasingly complex. Traditional power inspection methods mainly rely on manual inspection and regular maintenance. However, manual inspection has problems such as long inspection cycles, limited detection accuracy, and harsh inspection environments, making it difficult to meet the requirements of modern power systems for efficient and safe inspection. Therefore, intelligent inspection technology has gradually become a research hotspot, automatically identifying defects in power equipment through computer vision and deep learning technologies, improving inspection efficiency and detection accuracy, and reducing the workload and safety risks of manual inspection.

[0003] Currently, power equipment intelligent inspection systems mainly rely on object detection and image analysis technologies. By collecting visible light images and infrared thermal imaging data, surface defects, local temperature anomalies, discharge phenomena, etc. of the equipment are analyzed. Existing technologies usually use traditional object detection algorithms (such as Faster R-CNN, YOLO, etc.) for equipment status recognition. However, due to factors such as high-altitude operations, high-voltage environments, and complex background interference in transmission lines and substation equipment, traditional object detection algorithms have insufficient detection capabilities for small targets (such as insulator cracks and bolt looseness), prone to missed detections. In addition, the infrared thermal imaging analysis method usually uses a fixed threshold to determine equipment temperature anomalies, and fails to make adaptive adjustments in combination with the equipment operation environment and historical data, resulting in a high false alarm rate. In addition, existing systems lack effective trend analysis and warning mechanisms, making it difficult to accurately predict the evolution trend of equipment defects and affecting the early preventive maintenance of power equipment. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and device for image recognition safety warning of power equipment to solve the technical problem of insufficient detection capabilities of existing power equipment image recognition and detection methods for small targets.

[0005] To achieve the above object, the present invention adopts the following technical solutions: A method for image recognition safety warning of power equipment, comprising the following steps: Image acquisition and preprocessing; Based on an improved YOLO+Transformer model, perform object detection on the preprocessed image data; Based on the object detection results, use time series analysis methods to predict the evolution trend of defects; Set the warning level according to the prediction results, and trigger the corresponding warning response according to the warning level; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, add multi-head self-attention mechanism and sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; The K-Means++ algorithm is used to optimize the anchor box size, and the detection head is optimized in combination with the soft anchor allocation strategy.

[0006] Furthermore, the detection and warning results are uploaded to a remote monitoring platform, and further inspections are performed in conjunction with drones or inspection robots.

[0007] Furthermore, the image acquisition and preprocessing includes: Collect visible light images of power equipment and use infrared thermal imagers to obtain infrared thermal imaging data; The collected images are subjected to denoising, image enhancement, distortion correction and multi-scale feature extraction.

[0008] Furthermore, the time series analysis method uses a long short-term memory network to model the historical data of the target defect, analyze the defect change trend and predict future development.

[0009] Furthermore, the warning levels include low-level warning, intermediate warning and high-level warning. The low-level warning is used to record minor defects and arrange regular re-inspections. The intermediate warning recommends manual re-inspection. The high-level warning is used for emergency maintenance and links with intelligent inspection equipment for supplementary inspections.

[0010] Furthermore, the calculation formula for the target detection is:

[0011] In the formula, is the total loss function for target detection, Class classification loss, is the bounding box regression loss, is the IoU loss.

[0012] In a second aspect, the present invention provides a power equipment image recognition safety warning system, including an acquisition module, a detection module, a prediction module and a warning module, wherein: Acquisition module: used for image acquisition and preprocessing; Detection module: Based on the improved YOLO+Transformer model, target detection is performed on the preprocessed image data; Prediction module: Based on the target detection results, the evolution trend of defects is predicted using time series analysis methods; Early warning module: used to set the early warning level according to the prediction results, and trigger the corresponding early warning response according to the early warning level; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, add multi-head self-attention mechanism and sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; The K-Means++ algorithm is used to optimize the anchor box size, and the detection head is optimized in combination with the soft anchor allocation strategy.

[0013] Furthermore, it also includes a remote monitoring module for uploading the detection and warning results to the remote monitoring platform, and linking the intelligent inspection equipment to perform supplementary inspections.

[0014] According to a third aspect, a terminal device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0015] In a fourth aspect, a computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method are implemented.

[0016] Compared with the prior art, the present invention has the following beneficial technical effects: The invention discloses a safety warning method for image recognition of electric power equipment. The method performs target detection on preprocessed image data through an improved YOLO+Transformer model. The Transformer module is introduced and a multi-head self-attention mechanism is used to enhance the ability to distinguish between targets and backgrounds. The sliding window attention mechanism is combined to optimize the expression of local features, thereby improving the detection stability in complex environments. BiFPN feature fusion and DLA deep feature aggregation are used to optimize target detection, so that small target information can be effectively transmitted between feature layers of different scales, thereby reducing missed detection and false detection. In addition, K-Means++ is used to optimize the size of the Anchor frame, and the Soft Anchor Assignment strategy is combined to make target matching more accurate, thereby further improving the recall rate of small targets.

[0017] Preferably, the detection and early warning results are uploaded to a remote monitoring platform, enabling remote monitoring and online monitoring. In combination with drones or inspection robots for further inspections, the automation level of intelligent inspections is improved, the intelligent operation and maintenance of power equipment are achieved, the workload of manual inspections is effectively reduced, and the safety and maintenance efficiency of power grid operation are enhanced.

[0018] Preferably, the present invention also combines a time series analysis method to predict the evolution trend of equipment defects through an LSTM model, optimizing the early warning mechanism to make the safety early warning of power equipment more accurate and timely. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of a method for image recognition and safety early warning of power equipment in an embodiment of the present invention; Figure 2 It is a detailed flowchart diagram of a method for image recognition and safety early warning of power equipment in an embodiment of the present invention; Figure 3 It is a schematic diagram of the system module of a method for image recognition and safety early warning of power equipment in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] It should be noted that similar reference numerals and letters denote similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0022] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0023] Glossary: BiFPN: Bidirectional Feature Pyramid Network, an efficient multi-scale feature fusion architecture mainly used for object detection tasks. It optimizes feature interaction through bidirectional paths and a weighted mechanism, enhancing the model's detection ability for multi-scale objects.

[0024] DLA: Deep Layer Aggregation Network.

[0025] Anchor: Anchor box.

[0026] IoU: Intersection over Union, the intersection over union ratio, which is a core metric for measuring the overlap degree between two rectangular boxes or regions in computer vision tasks such as object detection and image segmentation.

[0027] Soft Anchor Assignment strategy: That is, the soft anchor assignment strategy.

[0028] The following further describes the present invention in detail with reference to the accompanying drawings: As Figure 1 shown, a method for image recognition and security warning of power equipment includes the following steps: Step 1, image acquisition and preprocessing; First, during the power equipment inspection process, a high-definition camera is used to obtain visible light images, and an infrared thermal imager is used to obtain the infrared thermal imaging data of the equipment to ensure comprehensive monitoring. Visible light images are mainly used to detect surface defects of power equipment, such as cracks, rust, loose bolts, etc., while infrared thermal imaging data is used to analyze temperature anomalies and identify potential hazards such as local overheating, discharge phenomena, or insulation faults.

[0029] The data acquisition device can be deployed at fixed monitoring points or on an unmanned aerial vehicle (UAV) inspection platform to periodically collect equipment operation status data. The obtained raw image data may contain problems such as noise, uneven illumination, and distortion, so data preprocessing is required to improve the accuracy of recognition.

[0030] Data preprocessing includes denoising, image enhancement, distortion correction, and multi-scale feature extraction to ensure the data quality input into the deep learning model.

[0031] Denoising: An adaptive filtering algorithm is used to remove environmental noise in the image and improve image clarity.

[0032] Image enhancement: Methods such as histogram equalization and gamma correction are applied to enhance the contrast of the image to improve the visibility of the defect area.

[0033] Distortion correction: Since UAV inspection may cause perspective distortion, perspective transformation and geometric correction methods are used to adjust the image to ensure the accuracy of the equipment form.

[0034] Multi-scale feature extraction: In infrared image processing, an adaptive temperature threshold segmentation method is adopted to extract the thermal distribution features of key components of the device and provide accurate input data for subsequent defect detection.

[0035] Step 2: Based on the improved YOLO+Transformer model, perform object detection on the preprocessed image data; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, and add the multi-head self-attention mechanism and the sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; Use the K-Means++ algorithm to optimize the Anchor size, and optimize the detection head in combination with the soft anchor assignment strategy.

[0036] As Figure 2 shown, first, multi-scale feature fusion uses BiFPN (bidirectional feature pyramid network) to enhance the feature expression ability of targets at different scales, enabling effective fusion of small target information such as insulator cracks and bolt looseness between different levels, and combining the deep feature aggregation (DLA) strategy to improve the detection accuracy of small targets.

[0037] Secondly, introduce the Transformer module into the YOLO backbone network, use the multi-head self-attention mechanism to optimize the target feature expression, enhance the discrimination ability between the target and the background, and at the same time use the sliding window attention mechanism to improve the local feature perception ability to ensure accurate identification of power equipment defects in complex backgrounds.

[0038] In addition, use the K-Means++ algorithm to optimize the Anchor box size to make the Anchor more match the size distribution of power equipment defect targets, and combine the Soft Anchor Assignment strategy to improve the target matching accuracy and enhance the recall rate of small targets.

[0039] The object detection calculation formula of this method is as follows:

[0040] In the formula, is the total loss function of object detection, is the class classification loss, is the bounding box regression loss, is the IoU loss.

[0041] Step 3: Based on the object detection results, use time series analysis method to predict the evolution trend of defects; Adopt a time series analysis model to predict the defect trend, combine historical defect data, and judge the evolution trend of equipment defects to provide early warnings.

[0042] The defect trend prediction is analyzed using a long short-term memory network (LSTM), and its core calculation logic is as follows:

[0043]

[0044] In the formula, is the hidden state at the current moment, is the defect feature input at the current moment, , , is the weight matrix obtained through training, , is the bias term, is the activation function, is the predicted defect severity.

[0045] This method combines the equipment historical defect data and the current detection data, analyzes the change trends of cracks, bolt looseness, or temperature anomalies through LSTM, and judges in advance whether the defects will deteriorate further to assist in the inspection and maintenance decisions.

[0046] Step 4: Set the warning level according to the prediction results, and trigger the corresponding warning response according to the warning level; As Figure 2 shown, based on the defect detection and trend prediction results, a three-level safety warning strategy is set to ensure the equipment safety; Low-level warning: Applicable to minor defects, such as slight rust and surface contamination. The system records the anomalies and reinspects during the next inspection.

[0047] Medium-level warning: Applicable to defects that may affect the equipment operation, such as bolt looseness and insulator surface damage. It is recommended to manually reinspect and evaluate the defect impact.

[0048] High-level warning: Applicable to serious defects, such as insulator rupture, severe discharge, and local overheating. The system immediately notifies the operation and maintenance personnel and initiates emergency repairs.

[0049] In an embodiment of the present invention, after the warning is triggered, the system can link the inspection drone for reinspection and send the detailed detection data to the remote monitoring system so that the operation and maintenance personnel can take corresponding measures in a timely manner.

[0050] In summary, the present invention provides a safety warning method for power equipment image recognition based on YOLO+Transformer, which combines multimodal data analysis, small target optimization detection, time series trend prediction and hierarchical warning mechanism to achieve efficient recognition and intelligent warning of power equipment defects. This method can be applied to the inspection tasks of transmission lines, substations and distribution equipment, reduce the workload of manual inspections, improve the intelligent level of equipment operation and maintenance, and ensure the safe and stable operation of the power system.

[0051] In yet another embodiment of the present invention, there is provided a power equipment image recognition safety warning system, comprising an acquisition module, a detection module, a prediction module and a warning module, wherein: Acquisition module: used for image acquisition and preprocessing; Detection module: Based on the improved YOLO+Transformer model, target detection is performed on the preprocessed image data; Prediction module: Based on the target detection results, the evolution trend of defects is predicted using time series analysis methods; Early warning module: used to set the early warning level according to the prediction results, and trigger the corresponding early warning response according to the early warning level; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, add multi-head self-attention mechanism and sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; The K-Means++ algorithm is used to optimize the anchor box size, and the detection head is optimized in combination with the soft anchor allocation strategy.

[0052] It also includes a remote monitoring module, which is used to upload detection and warning results to the remote monitoring platform and link intelligent inspection equipment to conduct supplementary inspections.

[0053] Optional, such as Figure 3 As shown, the system consists of multiple modules, including image acquisition module, data processing module, target recognition module, abnormal analysis module, safety warning module and remote monitoring module. Among them: Image acquisition module: used to obtain visible light images and infrared thermal imaging data of power equipment to ensure comprehensive monitoring of equipment defects.

[0054] Data processing module: performs pre-processing such as denoising, enhancement, and distortion correction on the collected image data to improve data quality and provide accurate input for subsequent target recognition.

[0055] Target recognition module: The improved YOLO+Transformer model is used for defect recognition to detect defects such as insulator cracks, loose bolts, partial discharge marks, and abnormal equipment temperature.

[0056] Abnormal analysis module: Based on time series analysis methods and combined with historical data, it predicts the development trend of defects and evaluates the possibility of failure.

[0057] Safety warning module: According to the equipment status and trend analysis results, set a three-level warning strategy, and combine the operation and maintenance scheduling mechanism to optimize the maintenance plan.

[0058] Remote monitoring module: uploads warning information to the cloud for remote monitoring by operation and maintenance personnel, and can link drones or inspection robots to perform supplementary inspection tasks.

[0059] In summary, the power equipment image recognition safety warning system of the present invention realizes efficient and accurate power equipment defect detection and warning through improved YOLO+Transformer target detection, time series analysis and intelligent warning strategy. The device can be widely used in the inspection of transmission lines, substations and distribution equipment, improve the intelligent level of power grid operation and maintenance, and ensure the safe and stable operation of the power system.

[0060] Application Examples This application example targets the transmission line inspection scenario and utilizes the power equipment image recognition safety warning method of the present invention to conduct intelligent inspections of key components such as transmission towers, insulators, and conductors to promptly detect equipment defects and improve the safety and stability of power grid operation.

[0061] During the inspection of power transmission lines, a combination of drone inspection and fixed monitoring equipment is usually used to achieve efficient equipment status detection. The drone is equipped with a high-definition camera and an infrared thermal imager, flying on a preset inspection path to collect image data of transmission towers and wires in real time. At the same time, the fixed monitoring equipment installed on the transmission tower continuously collects local features to supplement the lack of timeliness of drone inspections.

[0062] Data collection and preprocessing: During the transmission line inspection mission, the drone flies according to the planned route and uses a high-definition camera to collect images of the transmission tower's appearance to detect whether the transmission tower has rust, cracks, loose bolts and other problems. At the same time, the infrared thermal imager performs infrared inspection on the transmission line and insulator to identify potential risks such as overheating and discharge. Fixed monitoring equipment on the transmission tower also regularly collects relevant data to improve the continuity of system inspections.

[0063] The data collected by the monitoring device is denoised, enhanced, and distortion-corrected by the data processing module to remove background interference, improve data quality, and ensure the accuracy of subsequent deep learning model processing. After the data processing is completed, the system inputs the preprocessed images into the improved YOLO+Transformer model for defect recognition.

[0064] Defect detection and recognition: In this example, an improved YOLO+Transformer model is used for defect recognition. First, BiFPN (Bidirectional Feature Pyramid Network) is used to optimize the feature expression ability of targets at different scales, enabling effective fusion of small target information such as insulator cracks and bolt looseness between different levels. Then, a Transformer module is introduced into the backbone network of the YOLO network. The multi-head self-attention mechanism is used to enhance the discrimination ability between targets and the background, and the sliding window attention mechanism is combined to improve the perception ability of local features, thereby improving the detection accuracy of the model in complex environments.

[0065] During the target matching process, the system uses the K-Means++ algorithm to optimize the Anchor box size, making the target detection ability of the model more stable on different types of devices. At the same time, combined with the Soft Anchor Assignment strategy, the recall rate of small targets is improved to ensure accurate identification of tiny defects even in complex backgrounds.

[0066] For infrared image analysis, the system adopts an abnormal temperature distribution detection strategy, calculates the temperature distribution characteristics of the device, and dynamically adjusts the temperature threshold based on factors such as historical data, ambient temperature, and operating load to improve the accuracy of judgment. When the temperature of a certain part of the device is abnormally higher than the set threshold and this abnormality shows a trend of gradual expansion, the system determines that there may be device defects such as poor contact, partial discharge, or overload, and further analyzes the type of fault.

[0067] Trend prediction and safety warning: Based on the detected defect information, the system uses a time series analysis model to model the historical change trend of device defects and predict possible future fault situations. Different from traditional prediction methods based on a single time series, the present invention combines multi-modal data inputs, including visible light images, infrared thermal imaging data, and defect detection results, making the prediction results more in line with the actual operating state of the device.

[0068] During the trend prediction process, the system calculates the defect change rate and severity based on historical data, and combines a hierarchical warning mechanism to intelligently evaluate the device status. The warning strategy is as follows: Low-level warning: If there is slight rust on the surface of the transmission tower, the system records the abnormality and arranges subsequent inspections.

[0069] Medium-level warning: If small cracks or loose bolts are detected in the insulator, the system prompts the operation and maintenance personnel to arrange for manual reinspection.

[0070] High-level warning: If the abnormal increase in the wire temperature or the rupture of the insulator is found, the system immediately notifies the operation and maintenance personnel and initiates the emergency repair procedure.

[0071] Linkage between remote monitoring and intelligent inspection: All detection results are uploaded to the remote monitoring platform. The operation and maintenance personnel can view the device status through computer terminals or mobile APPs, and schedule drones for reinspection under the inspection plan recommended by the system, or arrange intelligent inspection robots for ground reinspection. When a high-level warning is detected, the system can automatically generate a maintenance report to assist the operation and maintenance personnel in troubleshooting.

[0072] This application example shows that the power equipment image recognition security warning method and device can be effectively applied to the inspection of transmission lines. Under complex working conditions such as high altitude, high voltage, and harsh environments, through drones and intelligent inspection equipment, intelligent defect detection and security warning of transmission equipment can be realized. This system improves the automation level of power inspection work through deep learning target detection technology, time series analysis, and intelligent warning strategies, reduces the risks and work intensity of manual inspection, and ensures the safe and stable operation of the power grid.

[0073] This example is also applicable to the intelligent operation and maintenance of power facilities such as substations and distribution equipment, and can provide technical support for the automatic inspection and fault warning of the power system.

[0074] Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0075] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, and the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, so that the instructions executed by the processors of the computer or other programmable data processing devices generate for implementation in the process Figure 1 a single process or multiple processes and / or blocksFigure 1 means for the functions specified in one or more boxes.

[0076] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction means that implements the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 or more boxes.

[0077] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus, such that a series of operational steps are performed on the computer or other programmable apparatus to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in one Figure 1 process or more processes and / or boxes Figure 1 or more boxes.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent replacements to the specific implementation manners of the invention, but these changes, modifications or equivalent replacements are all within the scope of protection of the pending claims of the invention.

Claims

1. A method for safety early warning of power equipment image recognition, characterized in that: The following steps are involved: Image acquisition and preprocessing; Based on the improved YOLO+Transformer model, target detection is performed on the preprocessed image data; Based on the target detection results, the evolution trend of defects is predicted using time series analysis methods; Set the warning level according to the prediction results, and trigger the corresponding warning response according to the warning level; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, add multi-head self-attention mechanism and sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; The K-Means++ algorithm is used to optimize the Anchor size, and the detection head is optimized in combination with the soft anchor allocation strategy.

2. According to claim 1, a method for safety early warning of power equipment image recognition is characterized in that: The detection and warning results are uploaded to the remote monitoring platform, and further inspections are performed in conjunction with drones or inspection robots.

3. The method for safety early warning of power equipment image recognition according to claim 1 is characterized in that: The image acquisition and preprocessing include: Collect visible light images of power equipment and use infrared thermal imagers to obtain infrared thermal imaging data; The collected images are subjected to denoising, image enhancement, distortion correction and multi-scale feature extraction.

4. The method for safety early warning of power equipment image recognition according to claim 1, characterized in that: The time series analysis method uses a long short-term memory network to model the historical data of target defects, analyze the defect change trend and predict future development.

5. The method for safety early warning of power equipment image recognition according to claim 1, characterized in that: The warning levels include low-level warning, intermediate warning and high-level warning. The low-level warning is used to record minor defects and arrange regular re-inspections. The intermediate warning recommends manual re-inspection. The high-level warning is used for emergency maintenance and links with intelligent inspection equipment for supplementary inspections.

6. The method for safety early warning of power equipment image recognition according to claim 1, characterized in that: The calculation formula for the target detection is: In the formula, is the total loss function for target detection, Class classification loss, is the bounding box regression loss, is the IoU loss.

7. An image recognition safety warning system for electric power equipment, characterized in that: A method for safety early warning of electric power equipment image recognition according to any one of claims 1 to 6 comprises an acquisition module, a detection module, a prediction module and an early warning module, wherein: Acquisition module: used for image acquisition and preprocessing; Detection module: Based on the improved YOLO+Transformer model, target detection is performed on the preprocessed image data; Prediction module: Based on the target detection results, the evolution trend of defects is predicted using time series analysis methods; Early warning module: used to set the early warning level according to the prediction results, and trigger the corresponding early warning response according to the early warning level; The improved YOLO+Transformer model includes: Embed the Transformer module in the YOLO backbone network, add multi-head self-attention mechanism and sliding window attention mechanism; Construct a bidirectional feature pyramid network and introduce a deep feature aggregation strategy; The K-Means++ algorithm is used to optimize the anchor box size, and the detection head is optimized in combination with the soft anchor allocation strategy.

8. The electric power equipment image recognition safety warning system according to claim 7, characterized in that: It also includes a remote monitoring module, which is used to upload detection and warning results to the remote monitoring platform and link intelligent inspection equipment to conduct supplementary inspections.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium storing 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