Target detection method and system for power equipment

Abstract

The application relates to the technical field of computer vision and target detection, and provides a target detection method and system for power equipment, which comprises the following steps: frame sampling is performed on a video stream; a frame image is converted into a gray image; feature points in the gray image are extracted by using a SURF algorithm; a single-channel region weight graph with the same size as the original gray image is generated; a region weighted image is obtained from the single-channel region weight graph; the region weighted image is adjusted to form a detection image meeting input requirements of YOLOv8; the detection image is input into a pre-trained YOLOv8 target detection network to generate candidate bounding boxes, confidence and category information; post-processing is performed on the output candidate bounding boxes, and the confidence is adjusted, so that a final detection result is output. The SURF is combined with the YOLOv8 deep detection, the data after the SURF pretreatment is used for training and detection at the same time, the end-to-end training of the network is not interfered, and the accuracy of target detection is improved.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and target detection technology, and in particular to a target detection method and system for power equipment. Background Technology

[0002] With the continuous expansion of modern power systems and the increasing level of automation, the safe operation of substations, as key power grid hubs, is of paramount importance. Traditional manual inspections suffer from drawbacks such as low efficiency, poor real-time performance, and susceptibility to human interference, making it difficult to meet the high requirements of modern substations for equipment status monitoring, fault early warning, and personnel safety monitoring. Automated monitoring systems based on video streams, utilizing deep learning object detection algorithms, have achieved some application in industrial monitoring, with the YOLO series being particularly favored for its end-to-end real-time detection capabilities.

[0003] However, in the complex environment of substations, there is often a large amount of irrelevant background information and overlapping targets between devices. Traditional YOLO detection methods are prone to missed detections or false detections in such scenarios. How to highlight the target area without significantly increasing the complexity of the network structure, thereby improving the detection accuracy, has become an urgent technical problem to be solved.

[0004] In existing technologies, some methods improve the network by introducing attention mechanisms and feature fusion, but these methods often increase network parameters, prolong training time, and may interfere with end-to-end convergence. On the other hand, SURF, as a classic hand-crafted feature extraction algorithm, can efficiently extract local salient features in images, but its traditional applications are mostly for static image feature matching and have not yet been organically combined with deep object detection networks. Summary of the Invention

[0005] The purpose of this invention is to solve at least one technical problem in the background art and to provide a target detection method and system for power equipment.

[0006] To achieve the above objectives, the present invention provides a target detection method for power equipment, comprising: Adaptive frame sampling is performed on the input video stream obtained from the substation monitoring system; Convert the frame image to grayscale and then reduce the noise in the grayscale image; The SURF algorithm is used to extract locally significant feature points in grayscale images. Based on the distribution of the extracted feature points in the grayscale image, a single-channel region weight map with the same size as the original grayscale image is generated. The single-channel region weight map is expanded into a three-channel map and multiplied pixel by pixel with the original frame image to obtain a region-weighted image. The size of the region-weighted image is adjusted to maintain the aspect ratio of the original image, and blank areas are filled to form a detection image that meets the input requirements of YOLOv8; The detected image is input into a pre-trained YOLOv8 object detection network, which generates candidate bounding boxes, confidence scores, and category information through forward propagation. The candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result.

[0007] According to one aspect of the present invention, during the generation of the region weight map, each feature point detected by SURF is locally diffused and accumulated using a 7×7 Gaussian kernel with a standard deviation of 2, and the resulting region weight map is linearly normalized so that its weight value is between 0.8 and 1.2.

[0008] According to one aspect of the present invention, the resizing of the region-weighted image, maintaining the original aspect ratio, and filling blank areas to form a detection image that meets the YOLOv8 input requirements is as follows: The region-weighted image is resized to 640×640 pixels using the Letterbox method, maintaining the original aspect ratio, and the blank areas are filled with RGB(114,114,114) to form a detection image that meets the input requirements of YOLOv8.

[0009] According to one aspect of the present invention, the candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result as follows: The candidate bounding boxes output by the YOLOv8 network are post-processed using the Soft-NMS algorithm. In the post-processing stage, the Soft-NMS algorithm uses a Gaussian decay function to adjust the confidence of the candidate bounding boxes, thereby outputting the final detection result.

[0010] According to one aspect of the present invention, the adaptive frame sampling of the input video stream acquired from the substation monitoring system includes: Convert the current sampled frame image and the previous sampled frame image into grayscale images; The difference between two frames of images was calculated using a normalized cross-correlation algorithm, with a threshold of 0.95 set. A sliding window mechanism is used to divide the frame image into a 5×5 grid. When more than 20% of the grid area has an NCC value below the threshold of 0.95, the current frame image is immediately processed and the sampling interval is reset to 1 frame. When the NCC value is higher than the threshold of 0.95, the sampling interval is increased by 2 times and the sampling interval is ≤30 frames.

[0011] To achieve the above objectives, the present invention also provides a target detection system for power equipment, comprising: The frame image sampling module performs adaptive frame sampling on the input video stream obtained from the substation monitoring system; The image conversion module converts the frame image into a grayscale image and reduces noise in the grayscale image; The feature point extraction module uses the SURF algorithm to extract locally significant feature points in the grayscale image; The single-channel region weight map generation module generates a single-channel region weight map with the same size as the original grayscale image based on the distribution of the extracted feature points in the grayscale image. The region-weighted image acquisition module expands the single-channel region weight map into a three-channel image and multiplies it pixel by pixel with the original frame image to obtain the region-weighted image. The detection image forming module resizes the region-weighted image, maintains the aspect ratio of the original image, and fills in blank areas to form a detection image that meets the input requirements of YOLOv8. The image detection module inputs the detected image into a pre-trained YOLOv8 object detection network. The YOLOv8 object detection network generates candidate bounding boxes, confidence scores, and category information through forward propagation. The detection result output module performs post-processing on the candidate bounding boxes output by the YOLOv8 network and adjusts the confidence level to output the final detection result.

[0012] To achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the target detection method for power equipment as described above.

[0013] To achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the target detection method for power equipment as described above.

[0014] According to the present invention, adaptive frame sampling technology is introduced to effectively reduce device load and optimize system resource utilization. By applying the SURF algorithm as an independent preprocessing module to the video stream, local salient features are extracted and a region weight map is generated. This weight map is then used to perform region weighting processing on the video frames, making the weighted image more prominent in key target regions and reducing background interference. The processed image is then input into the YOLOv8 detection network for target detection. This scheme not only improves detection accuracy but also, because the SURF module does not participate in backpropagation, does not affect the training convergence of the YOLOv8 network.

[0015] This invention uses the YOLO algorithm. While traditionally the original image is directly input into the network, this invention introduces SURF to generate region weight maps before using the YOLO algorithm. This region weighting highlights key target regions, thereby improving detection accuracy. The SURF module is an independent preprocessing step, not involved in network training, ensuring real-time performance and preventing interference with end-to-end YOLOv8 training. Simultaneously, Gaussian kernel diffusion and normalization achieve simple and efficient region weighting. In the post-processing stage, the Soft-NMS algorithm is used to further optimize candidate box selection, further improving detection performance. Attached Figure Description

[0016] Figure 1 The flowchart schematically illustrates a target detection method for power equipment according to one embodiment of the present invention. Detailed Implementation

[0017] The invention will now be discussed with reference to exemplary embodiments. It should be understood that the described embodiments are merely intended to enable those skilled in the art to better understand and thus implement the invention, and are not intended to imply any limitation on the scope of the invention.

[0018] As used herein, the term "comprising" and its variations are to be interpreted as open-ended terms meaning "including but not limited to". The term "based on" is to be interpreted as "at least partially based on". The terms "one embodiment" and "an embodiment" are to be interpreted as "at least one embodiment".

[0019] Figure 1 A flowchart illustrating a target detection method for power equipment according to an embodiment of the present invention is shown. Figure 1 As shown, in this embodiment, the target detection method for power equipment includes: Adaptive frame sampling is performed on the input video stream obtained from the substation monitoring system; Convert the frame image to grayscale and then reduce the noise in the grayscale image; The SURF algorithm is used to extract locally significant feature points in grayscale images. Based on the distribution of the extracted feature points in the grayscale image, a single-channel region weight map with the same size as the original grayscale image is generated. The single-channel region weight map is expanded into a three-channel map and multiplied pixel by pixel with the original frame image to obtain a region-weighted image. The size of the region-weighted image is adjusted to maintain the aspect ratio of the original image, and blank areas are filled to form a detection image that meets the input requirements of YOLOv8; The detected image is input into a pre-trained YOLOv8 object detection network, which generates candidate bounding boxes, confidence scores, and category information through forward propagation. The candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result.

[0020] Furthermore, according to one embodiment of the present invention, during the generation of the region weight map, each feature point detected by SURF is locally diffused and accumulated using a 7×7 Gaussian kernel with a standard deviation of 2, and the resulting region weight map is linearly normalized so that its weight value is between 0.8 and 1.2.

[0021] Furthermore, according to one embodiment of the present invention, the size of the region-weighted image is adjusted while maintaining the original aspect ratio, and blank areas are filled to form a detection image that meets the input requirements of YOLOv8: The region-weighted image is resized to 640×640 pixels using the Letterbox method, maintaining the original aspect ratio, and the blank areas are filled with RGB(114,114,114) to form a detection image that meets the input requirements of YOLOv8.

[0022] Furthermore, according to one embodiment of the present invention, the candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result as follows: The candidate bounding boxes output by the YOLOv8 network are post-processed using the Soft-NMS algorithm. In the post-processing stage, the Soft-NMS algorithm uses a Gaussian decay function to adjust the confidence of the candidate bounding boxes, with parameters of IoU threshold of 0.6 and sigma parameter of 0.5, thereby outputting the final detection result.

[0023] Furthermore, according to one embodiment of the present invention, adaptive frame sampling is performed on the input video stream obtained from the substation monitoring system, including: Convert the current sampled frame image and the previous sampled frame image into grayscale images; The difference between two frames of images was calculated using a normalized cross-correlation algorithm, with a threshold of 0.95 set. A sliding window mechanism is used to divide the frame image into a 5×5 grid. When more than 20% of the grid area has an NCC value below the threshold of 0.95, the current frame image is immediately processed and the sampling interval is reset to 1 frame. When the NCC value is higher than the threshold of 0.95, the sampling interval is increased by 2 times and the sampling interval is ≤30 frames.

[0024] According to the above-described scheme of the present invention, the present invention effectively reduces device load and optimizes system resource utilization by introducing adaptive frame sampling technology. By applying the SURF algorithm as an independent preprocessing module to the video stream, local salient features are extracted and a region weight map is generated. This weight map is then used to perform region weighting processing on the video frames, making the weighted image more prominent in key target areas and reducing background interference. The processed image is then input into the YOLOv8 detection network for target detection. This scheme not only improves detection accuracy, but also, because the SURF module does not participate in backpropagation, it does not affect the training convergence of the YOLOv8 network.

[0025] This invention uses the YOLO algorithm. While traditionally the original image is directly input into the network, this invention introduces SURF to generate region weight maps before using the YOLO algorithm. This region weighting highlights key target regions, thereby improving detection accuracy. The SURF module is an independent preprocessing step, not involved in network training, ensuring real-time performance and preventing interference with end-to-end YOLOv8 training. Simultaneously, Gaussian kernel diffusion and normalization achieve simple and efficient region weighting. In the post-processing stage, the Soft-NMS algorithm is used to further optimize candidate box selection, further improving detection performance.

[0026] Furthermore, to achieve the above objectives, the present invention also provides a target detection system for power equipment, comprising: The frame image sampling module performs adaptive frame sampling on the input video stream obtained from the substation monitoring system; The image conversion module converts the frame image into a grayscale image and reduces noise in the grayscale image; The feature point extraction module uses the SURF algorithm to extract locally significant feature points in the grayscale image; The single-channel region weight map generation module generates a single-channel region weight map with the same size as the original grayscale image based on the distribution of the extracted feature points in the grayscale image. The region-weighted image acquisition module expands the single-channel region weight map into a three-channel image and multiplies it pixel by pixel with the original frame image to obtain the region-weighted image. The detection image forming module resizes the region-weighted image, maintains the aspect ratio of the original image, and fills in blank areas to form a detection image that meets the input requirements of YOLOv8. The image detection module inputs the detected image into a pre-trained YOLOv8 object detection network. The YOLOv8 object detection network generates candidate bounding boxes, confidence scores, and category information through forward propagation. The detection result output module performs post-processing on the candidate bounding boxes output by the YOLOv8 network and adjusts the confidence level to output the final detection result.

[0027] The target detection system for power equipment described above according to the present invention can realize the target detection method for power equipment described above. The specific process steps are as described above and will not be repeated here.

[0028] Furthermore, to achieve the above objectives, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the target detection method for power equipment as described above.

[0029] Furthermore, to achieve the above objectives, the present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the target detection method for power equipment as described above.

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely one preferred embodiment of the invention and are only used to explain the invention. They do not limit the scope of protection of the invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0031] Example 1 The target detection method for power equipment includes the following steps: 1. Adaptive frame sampling and resource optimization: To address the long-term stable equipment status in substation scenarios, a dynamic sampling strategy and a dual-mode detection mechanism are introduced at the video stream input: combining global NCC similarity (threshold 0.95) with local 5×5 gridded detection (an alarm is triggered if the NCC of a single grid is <0.9). When the global similarity is high and there are no local anomalies, the sampling interval gradually increases from 1 frame to 30 frames (a doubling increment). Emergency response rule: When a change in equipment status or a fault is detected, the system immediately enters frame-by-frame processing mode.

[0032] 2. SURF Feature Extraction and Scene Adaptation: The RGB image is converted to grayscale, and noise is reduced using a 3×3 Gaussian blur filter. The SURF algorithm is then used to extract local salient feature points at multiple scales. Parameter settings: Hessian threshold 500, number of octaves 4, number of layers per octave 3. This step outputs the coordinates of all key feature points in the image.

[0033] 3. Generate a region weight map: Based on the key feature points extracted by SURF, an accumulation process is performed on a single-channel image of the same size as the original image. For each key feature point, a 7×7 Gaussian kernel with a standard deviation of 2 is used for local diffusion accumulation at its location to form a regional heatmap. The accumulation result is then linearly normalized, and the normalization formula can be set as: ; in, and These represent the minimum and maximum values ​​of the weighted map, respectively. The normalized weight values ​​are limited to between 0.8 and 1.2, allowing regions with more feature points to receive higher weights.

[0034] 4. Generate region-weighted images: The single-channel region weight map is expanded into a three-channel image and multiplied pixel-by-pixel with the original RGB image to obtain a region-weighted image. In this step, the pixel values ​​of the target region are enhanced by multiplying by a weight greater than 1, while the background region is weakened due to its lower weight. For example, in the resulting image, equipment such as switches and transformers in a substation receive higher weights due to their rich key points, resulting in enhanced brightness; while the background region is relatively weakened.

[0035] Subsequently, the weighted image is normalized to ensure that all pixel values ​​are within the range of 0 to 1, thus avoiding the introduction of abnormal values ​​due to the weighting operation.

[0036] 5. Letterbox image adjustment: The weighted image was resized using the Letterbox method. The target size was preset to 640×640 pixels. To maintain the original aspect ratio, the fill color for any insufficient areas was set to RGB(114,114,114). Then, interpolation was used for scaling to ensure sharp details on the target device. This step ensured that the image information was not distorted and met the input requirements of the YOLOv8 network.

[0037] 6. YOLOv8 Object Detection: The region-weighted and letterbox-adjusted image is fed as input into the pre-trained or fine-tuned YOLOv8 object detection network. The YOLOv8 network utilizes deep convolutional structures for multi-scale feature extraction, generating candidate bounding boxes and corresponding object categories and confidence scores. Since specific equipment and personnel targets exist in substation scenarios, samples containing substation monitoring data can be used for fine-tuning during the training phase to make the network more adaptable to the field environment. Target categories include switchgear, faulty equipment, and personnel.

[0038] 7. Post-processing - Soft-NMS: Post-processing is performed on the candidate bounding boxes output by the YOLOv8 network, using the Soft-NMS algorithm to eliminate redundant boxes. The Soft-NMS algorithm employs a Gaussian decay function, attenuating the score of a candidate bounding box based on the Intersection over Union (IoU) between it and the highest confidence box, rather than directly deleting it. The IoU threshold is set to 0.6, and the sigma parameter is set to 0.5.

[0039] 8. Detection result output and time-domain fusion: After Soft-NMS processing, the final detection results are output, including candidate box coordinates, target category, and confidence level. The detection results are applied to real-time monitoring, displayed promptly on the monitoring system interface, or used to trigger alarm signals. To improve the stability of continuous video stream detection, detection results in consecutive frames are fused in the temporal domain using Kalman filtering, further reducing false positive and false negative rates and ensuring the continuity and reliability of monitoring results.

[0040] Regional weighted preprocessing enhances information about critical equipment areas and significantly reduces background noise interference, making it easier for YOLOv8 networks to detect critical targets in substations, such as switchgear, faulty components, and abnormal personnel behavior. The accuracy of switchgear status monitoring, equipment fault detection, and personnel safety monitoring is improved by approximately 5%–10% compared to traditional YOLOv8 solutions, while the false alarm rate is reduced by approximately 10%.

[0041] Those skilled in the art will recognize that the modules and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0042] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and equipment can be referred to the corresponding process in the foregoing method implementation, and will not be repeated here.

[0043] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0044] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the objectives of the embodiments of the present invention, depending on actual needs.

[0045] In addition, the functional modules in the embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0046] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the sending / receiving methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0047] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in this application.

[0048] It should be understood that the sequence number of each step in the invention and its embodiments does not absolutely imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

Claims

1. A target detection method for power equipment, characterized in that, include: Adaptive frame sampling is performed on the input video stream obtained from the substation monitoring system, including: converting the current sampled frame image and the previous sampled frame image into grayscale images; calculating the difference between the two frames using a normalized cross-correlation algorithm and setting a threshold of 0.95; dividing the frame image into a 5×5 grid using a sliding window mechanism; when more than 20% of the grid area has an NCC value lower than the threshold of 0.95, the current frame image is immediately processed and the sampling interval is reset to 1 frame; when the NCC value is higher than the threshold of 0.95, the sampling interval increases by a factor of 2, and the sampling interval is ≤30 frames. Denoising grayscale images; The SURF algorithm is used to extract locally significant feature points in grayscale images. Based on the distribution of the extracted feature points in the grayscale image, a single-channel region weight map with the same size as the original grayscale image is generated. This includes: accumulating the key feature points extracted using the SURF algorithm on a single-channel image with the same size as the original image; for each key feature point, performing local diffusion accumulation at its location using a 7×7 Gaussian kernel with a standard deviation of 2 to form a region heatmap; and linearly normalizing the accumulation result. The normalization formula can be set as: ； in, and These are the minimum and maximum values ​​of the weighted map, respectively; the normalized weight values ​​are limited to between 0.8 and 1.2, so that regions with more feature points receive higher weights; The single-channel region weight map is expanded into a three-channel map and multiplied pixel by pixel with the original frame image to obtain a region-weighted image. The size of the region-weighted image is adjusted to maintain the aspect ratio of the original image, and blank areas are filled to form a detection image that meets the input requirements of YOLOv8; The detected image is input into a pre-trained YOLOv8 object detection network, which generates candidate bounding boxes, confidence scores, and category information through forward propagation. The candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result.

2. The target detection method for power equipment according to claim 1, characterized in that, During the generation of the single-channel region weight map, each feature point detected by SURF is locally diffused and accumulated using a 7×7 Gaussian kernel with a standard deviation of 2, and the resulting region weight map is linearly normalized to make its weight value between 0.8 and 1.

2.

3. The target detection method for power equipment according to claim 1, characterized in that, The process of resizing the region-weighted image, maintaining the original aspect ratio, and filling in blank areas to form a detection image that meets the input requirements of YOLOv8 is as follows: The region-weighted image is resized to 640×640 pixels using the Letterbox method, maintaining the original aspect ratio, and the blank areas are filled with RGB(114,114,114) to form a detection image that meets the input requirements of YOLOv8.

4. The target detection method for power equipment according to claim 1, characterized in that, The candidate bounding boxes output by the YOLOv8 network are post-processed and the confidence level is adjusted to output the final detection result: The candidate bounding boxes output by the YOLOv8 network are post-processed using the Soft-NMS algorithm. In the post-processing stage, the Soft-NMS algorithm uses a Gaussian decay function to adjust the confidence of the candidate bounding boxes, thereby outputting the final detection result.

5. A target detection system for power equipment, characterized in that, include: The frame image sampling module performs adaptive frame sampling on the input video stream acquired from the substation monitoring system, including: converting the current sampled frame image and the previous sampled frame image into grayscale images; calculating the difference between the two frames using a normalized cross-correlation algorithm and setting a threshold of 0.95; dividing the frame image into a 5×5 grid using a sliding window mechanism; when more than 20% of the grid area has an NCC value lower than the threshold of 0.95, immediately processing the current frame image and resetting the sampling interval to 1 frame; and when the NCC value is higher than the threshold of 0.95, the sampling interval increases by a factor of 2, and the sampling interval is ≤30 frames. The image conversion module reduces noise in grayscale images; The feature point extraction module uses the SURF algorithm to extract locally significant feature points in the grayscale image; The single-channel region weight map generation module generates a single-channel region weight map of the same size as the original grayscale image based on the distribution of extracted feature points in the grayscale image. This includes: accumulating key feature points extracted using the SURF algorithm on a single-channel image of the same size as the original image; performing local diffusion accumulation at the location of each key feature point using a 7×7 Gaussian kernel with a standard deviation of 2 to form a region heatmap; and linearly normalizing the accumulation result using a formula that can be set as follows: ； in, and These are the minimum and maximum values ​​of the weighted map, respectively; the normalized weight values ​​are limited to between 0.8 and 1.2, so that regions with more feature points receive higher weights; The region-weighted image acquisition module expands the single-channel region weight map into a three-channel image and multiplies it pixel by pixel with the original frame image to obtain the region-weighted image. The detection image forming module resizes the region-weighted image, maintains the aspect ratio of the original image, and fills in blank areas to form a detection image that meets the input requirements of YOLOv8. The image detection module inputs the detected image into a pre-trained YOLOv8 object detection network. The YOLOv8 object detection network generates candidate bounding boxes, confidence scores, and category information through forward propagation. The detection result output module performs post-processing on the candidate bounding boxes output by the YOLOv8 network and adjusts the confidence level to output the final detection result.

6. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the target detection method for power equipment as described in any one of claims 1-5.

7. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the target detection method for power equipment as described in any one of claims 1-5.

Images