A silica gel barrel discoloration detection method, device, equipment and storage medium

Abstract

This invention relates to the field of intelligent detection, and discloses a method, apparatus, device, and storage medium for detecting discoloration of silicone barrels. The method includes: acquiring an image of a respirator in a substation; performing target detection on the respirator image, selecting respirator images containing silicone barrels as target images, and cropping a silicone barrel region image from the target image; segmenting the silicone barrel region image horizontally according to a preset segmentation rule to obtain multiple local sub-images; extracting features from each local sub-image, comparing the extracted features with a preset feature library to obtain comparison results for each local sub-image; and calculating the degree of discoloration of the silicone barrel in the respirator based on the comparison results of each local sub-image. This invention solves the problem of low accuracy in detecting discoloration of silicone barrels in respirators in existing technologies.

Description

Technical Field

[0001] This invention relates to the field of intelligent detection, and in particular to a method, apparatus, equipment, and storage medium for detecting color change in silicone buckets. Background Technology

[0002] In existing technologies, the detection of discoloration of silicone barrels in substation breathers mostly relies on manual inspections or general image recognition solutions. Manual inspections are limited by subjective experience and inspection cycles, making them prone to missed detections and misjudgments. General image recognition solutions are not adapted to the complex outdoor environment of substations (such as backlight, haze, and dust interference) and the unique characteristic of moisture seeping from top to bottom into silicone barrels. They lack specificity in the segmentation and feature extraction of images of the silicone barrel area, resulting in low detection accuracy and an inability to reliably reflect the moisture status of the silicone, which makes it difficult to meet the safety assurance requirements of substation equipment operation and maintenance. Summary of the Invention

[0003] This invention provides a method, apparatus, computer equipment, and storage medium for detecting color change in silicone tubs, in order to solve the problem of low accuracy in detecting color change in respirator silicone tubs in the prior art.

[0004] Firstly, a method for detecting color change in silicone barrels is provided, including:

[0005] Acquire images of the respirator in the substation;

[0006] Target detection is performed on the respirator images, and images containing silicone barrels are selected as target images. The silicone barrel region is then extracted from the target images.

[0007] The image of the silicone bucket region is divided horizontally according to a preset segmentation rule to obtain multiple local sub-images;

[0008] Feature extraction is performed on each local sub-image, and the extracted features are compared with a preset feature library to obtain the comparison results of each local sub-image.

[0009] Based on the comparison results of each local sub-image, the degree of discoloration of the silicone barrel in the respirator is calculated.

[0010] Optionally, the image of the silicone bucket region is segmented horizontally according to a preset segmentation rule to obtain multiple local sub-images, including:

[0011] The image of the silicone bucket area is divided into an upper core detection area and a lower auxiliary detection area according to a preset division rule;

[0012] According to the preset fine segmentation rules, the upper core detection area is divided into multiple local sub-images along the horizontal direction;

[0013] According to the preset coarse segmentation rules, the lower auxiliary detection area is divided into multiple local sub-images along the horizontal direction;

[0014] The fine-grained segmentation rule has a smaller segmentation granularity than the coarse-grained segmentation rule.

[0015] Optional, also includes:

[0016] Calculate the environmental interference coefficient based on the image of the silicone bucket area;

[0017] If the environmental interference coefficient is greater than the threshold, then reduce the granularity of the fine segmentation rule and the coarse segmentation rule;

[0018] If the environmental interference coefficient is not greater than the threshold, the silicone bucket region image is segmented according to the conventional segmentation granularity of fine segmentation rules and coarse segmentation rules.

[0019] Optionally, feature extraction is performed on each local sub-image, and the extracted features are compared with a preset feature library to obtain the comparison results of each local sub-image, including:

[0020] Based on a pre-defined feature extraction network, features of local sub-images are extracted;

[0021] Based on the similarity algorithm, the similarity scores between the feature and the normal silicone feature and the color-changing silicone feature in the feature library are calculated respectively to obtain the normal similarity and the color-changing similarity.

[0022] The highest values ​​of the preset number are selected from both the normal similarity and the color-changing similarity, and the average similarity of the normal and color-changing samples are calculated respectively.

[0023] If the normal average similarity is greater than the color-changing average similarity, the comparison result of the local sub-image is normal; if the normal average similarity is not greater than the color-changing average similarity, the comparison result of the local sub-image is color-changing.

[0024] Optionally, based on the comparison results of each local sub-image, the degree of discoloration of the silicone canister in the respirator is calculated, including:

[0025] Obtain the comparison results of each local sub-image, including whether it is normal or discolored;

[0026] Obtain the area of ​​each local sub-image that changes color in the comparison result;

[0027] The ratio of the sum of the areas of the discolored local sub-images to the sum of the areas of all local sub-images is calculated, and this ratio is the degree of discoloration of the silicone tub in the respirator.

[0028] Optionally, the preset feature library is a multi-scene feature library, and its construction process includes:

[0029] Images of respirators in different scenarios were collected as sample images;

[0030] The sample images were filtered to obtain sample images containing silicone buckets;

[0031] The sample image of the silicone bucket is obtained by cropping the sample image containing the silicone bucket;

[0032] The sample silicone bucket image is segmented to obtain local sub-images of the sample;

[0033] Local sub-images of the sample are labeled as normal samples or discolored samples according to the actual state of the silicone.

[0034] By using a pre-defined feature extraction network, local sub-image features of normal or discolored samples are extracted, and then classified and stored according to different scenarios to obtain a multi-scenario feature library.

[0035] Feature extraction is performed on each local sub-image, and the extracted features are compared with a pre-defined feature library, including:

[0036] The target scene feature library matching the current substation is determined from the multi-scene feature library, and the extracted features are compared with the target scene feature library.

[0037] Optionally, after calculating the degree of discoloration of the silicone cartridge in the respirator, the following are included:

[0038] By comparing the degree of color change with the preset warning threshold, the state level of the silicone bucket can be obtained;

[0039] Status reports and early warning signals are generated based on the status level.

[0040] Secondly, a device for detecting color change in silicone barrels is provided, comprising:

[0041] The acquisition module is used to acquire images of the respirator in the substation;

[0042] The filtering module is used to perform target detection on respirator images, filter out respirator images containing silicone barrels as target images, and extract silicone barrel region images from target images.

[0043] The segmentation module is used to segment the image of the silicone bucket region along the horizontal direction according to a preset segmentation rule to obtain multiple local sub-images;

[0044] The comparison module is used to extract features from each local sub-image, compare the extracted features with a preset feature library, and obtain the comparison results of each local sub-image.

[0045] The calculation module is used to calculate the degree of discoloration of the silicone tubing in the respirator based on the comparison results of each local sub-image.

[0046] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described silicone bucket color change detection method.

[0047] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the aforementioned silicone bucket color change detection method.

[0048] The aforementioned method, apparatus, equipment, and storage medium for detecting discoloration of silicone barrels first screens out images containing silicone barrels through target detection and extracts the silicone barrel area, eliminating background interference to ensure that subsequent analysis focuses on the silicone barrel area. The core discoloration of silicone in substation breather respirators is "from top to bottom" (moisture and impurities first contact the upper silicone and gradually penetrate downwards). Horizontal segmentation accurately matches this non-uniform distribution characteristic, avoiding the problem of normal and discolored areas being mixed in the same local sub-image, ensuring that each sub-image reflects the true state of the corresponding horizontal area. Horizontal segmentation according to preset rules accurately captures discoloration details at different heights (such as slight discoloration in the upper core area and local discoloration in the lower auxiliary area). Finally, the discolored area stitched together from the sub-image states more closely matches the actual discoloration of the silicone barrel. By comparing the local sub-image features with a preset feature library, an objective and unified judgment standard is established, and the silicone state is quantified through feature matching. Therefore, this solves the problem of low accuracy in detecting discoloration of respirator silicone barrels in existing technologies. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic flowchart of a method for detecting color change in a silicone bucket according to an embodiment of the present invention;

[0051] Figure 2 This is a schematic diagram of a silicone bucket color change detection device according to an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0053] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0054] It should also be understood that the term “and / or” as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0055] As used in this specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," or "in response to determination." Similarly, the phrase "if determined" or "if matched to [described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once matched to [described condition or event]," or "in response to matched to [described condition or event]."

[0056] Furthermore, in the description of this invention and the appended claims, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0057] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of the invention include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0058] It should be understood that the sequence number of each step in the following embodiments does not 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.

[0059] To illustrate the technical solution of the present invention, specific embodiments are described below.

[0060] Please see Figure 1 As shown, Figure 1 A schematic flowchart of a method for detecting color change in a silicone bucket provided in an embodiment of the present invention includes the following steps:

[0061] S11: Obtain an image of the respirator in the substation.

[0062] Since the feature extraction effect of RGB three-channel color images has been fully verified, it can accurately capture key features such as color differences and texture details of silicone in normal / discoloration states. Therefore, the image type acquired in this step is generally RGB color images, and the acquisition method needs to be achieved through the camera points preset in the substation (such as point-based PTZ cameras in the inspection system).

[0063] The images collected in this step need to cover the actual installation scenario of the breather in the substation, including different equipment models (such as main transformer breather and instrument transformer breather), installation height, placement angle, and different environmental conditions (such as sunny days, cloudy days, nighttime supplemental lighting, dust interference, etc.) to ensure that the collected images can reflect the status of the silicone barrel under real working conditions.

[0064] In one example, using the pan-tilt-zoom (PTZ) cameras already deployed in the substation and the high-definition camera mounted on the inspection robot, images of the respirator are automatically captured from the front or at a 45° angle (e.g., every 2 hours, capturing 3 frames of images with different exposures each time) according to a preset inspection route and shooting locations. For example, in the main transformer area of ​​a 220kV substation, an RGB color image is captured by aiming a pan-tilt-zoom camera (4K resolution, adjustable focus) at the silicone barrel of the respirator.

[0065] S12: Perform target detection on the respirator image, filter out the respirator image containing the silicone barrel as the target image, and extract the silicone barrel area image from the target image.

[0066] The core function of this step is to precisely isolate the silicone barrel area from complex respirator images.

[0067] Images of substation breather devices often contain redundant information such as equipment casing, wires, and background environment. Direct feature extraction can introduce interference, leading to a decrease in feature discrimination. This step, through target detection and region cropping, can eliminate redundant information, focusing on the core area of ​​the silicone drum. This avoids interference from non-silicone drum areas on subsequent feature comparison and color change calculation, ensuring detection accuracy.

[0068] Specifically, this can be achieved using the YOLOv8-Detect object detection model: First, train the model using labeled silicone bucket sample data to enable it to recognize the silicone bucket region in the ventilator image. Then, input the acquired ventilator image into the trained model, and the model outputs the target bounding box coordinates of the silicone bucket. If the silicone bucket is detected, the image is the target image, and the silicone bucket region image is cropped based on the target bounding box coordinates. If it is not detected, return to the image acquisition stage to reacquire the ventilator image.

[0069] In one example, a pan-tilt-zoom camera in a substation inspection system captures an image (RGB format) of a respirator and inputs it into a trained YOLOv8-Detect model. The model uses neural network inference to identify a silicone bucket in the image and outputs the target bounding box coordinates (x1=200, y1=150, x2=400, y2=500). This image is then identified as the target image. Subsequently, a rectangular region is cropped based on the coordinates (200≤x≤400, 150≤y≤500), resulting in an image containing only the silicone bucket, eliminating redundant parts such as the respirator casing and surrounding wires.

[0070] S13: Divide the image of the silicone bucket region along the horizontal direction according to the preset segmentation rules to obtain multiple local sub-images.

[0071] The core purpose of this step is to improve the granularity of the silicone color-changing area by segmentation, so as to make subsequent feature extraction and comparison more accurate and solve the problem of insufficient fine granularity caused by the existing target detection output rectangle.

[0072] The discoloration of silicone barrels is often non-uniform (e.g., localized areas change color first). If features are extracted directly from the entire area, the features of normal and discolored areas will interfere with each other, leading to distorted comparison results. However, after segmentation, each local sub-image has more singular features, which can accurately reflect the silicone state of the corresponding area, ensuring the accuracy of subsequent comparisons and discoloration degree calculations.

[0073] The segmentation rule typically involves uniformly segmenting the silicone bucket region image horizontally according to preset granularity parameters. This horizontally uniform segmentation method decomposes the entire silicone bucket region into multiple local sub-images, with each sub-image corresponding to a local area at a different horizontal position within the silicone bucket. This avoids the masking of local discoloration features due to overall analysis. Simultaneously, the segmented local sub-images can be individually feature-extracted, accurately capturing the color and texture details of each locality. This provides more discriminative feature data for subsequent comparison with a feature library, ultimately ensuring that the discoloration area closely matches reality and improving detection accuracy.

[0074] In one example, the horizontal segment height is set to 15 pixels per segment. If the horizontal height of the cropped silicone bucket area image is 300 pixels, after being uniformly segmented along the Y-axis, 20 local sub-images are obtained (each sub-image is sized as "silicone bucket horizontal width × 15 pixels"), completely covering the entire area of ​​the silicone bucket from top to bottom.

[0075] In another example, for different models of breathers in substations, the heights of the cropped silicone barrel area images are 240 pixels and 330 pixels, respectively. Following preset rules, a fixed number of 16 local sub-images is used as the standard, and the segmentation height is calculated separately (240÷16=15 pixels / segment, 330÷16≈20.6 pixels / segment), uniformly segmented along the Y-axis. This ensures a consistent number of sub-images while adapting to the detection needs of silicone barrels of different sizes.

[0076] S14: Extract features from each local sub-image, compare the extracted features with the preset feature library, and obtain the comparison results of each local sub-image.

[0077] The core purpose of this step is to clarify the silicone state (normal / discolored) of each local sub-image through precise feature extraction and comparison, providing a basis for the final calculation of the degree of discoloration.

[0078] This step is implemented in two steps: First, feature extraction, which typically uses a ResNet50 feature extraction network to extract features from each local sub-image. This network can accurately capture key features such as the color and texture of silicone in the local sub-image, forming a high-dimensional feature vector. Compared with traditional methods, it can better distinguish the subtle differences between normal and discolored silicone under different conditions. Second, feature comparison, using a preset feature library as a reference, and cosine similarity as a metric, determines the state of the local sub-image by calculating the similarity between the extracted features and the normal and discolored features in the feature library.

[0079] In one example, normal silicone rubber in a substation is blue, and color-changing silicone rubber is pink. After segmentation according to rules, a local sub-image A (silicone rubber in blue) is obtained. The feature vector of A is extracted using a ResNet50 network. The cosine similarity between A and the features of normal silicone rubber in the preset feature library is calculated to be 0.92, and the cosine similarity between A and the features of color-changing silicone rubber is 0.35. Since 0.92 > 0.35, the local sub-image A is determined to be "normal".

[0080] S15: Based on the comparison results of each local sub-image, calculate the degree of discoloration of the silicone barrel in the respirator.

[0081] This step is the core of the entire testing process, and its main purpose is to obtain the degree of discoloration of the silicone barrel through quantitative calculation, and to convert the qualitative judgment results (normal / discoloration) of each local sub-image into quantitative values.

[0082] In some embodiments, calculating the degree of discoloration of the silicone tub in the respirator based on the comparison results of each local sub-image includes: obtaining the comparison results of each local sub-image, the comparison results including normal or discolored; obtaining the area of ​​each local sub-image whose comparison result is discolored; and calculating the ratio of the sum of the areas of the discolored local sub-images to the sum of the areas of all local sub-images, the ratio being the degree of discoloration of the silicone tub in the respirator.

[0083] The calculation method in this embodiment is based on the premise that "local sub-images completely cover the silicone bucket region": all local sub-images are segments of the silicone bucket region image, and their total area is exactly the same as the area of ​​the silicone bucket region. Therefore, the ratio of "the sum of the areas of the color-changing sub-images ÷ the sum of the areas of all sub-images" can directly quantify the proportion of the color-changing region in the silicone bucket, accurately reflecting the degree of color change. This design not only connects with the results of the previous "local sub-image feature comparison" but also avoids subjective judgment errors, making the output results objective and repeatable, significantly improving detection accuracy. Moreover, this calculation method does not require complex algorithms; it can be achieved simply through area statistics and ratio calculation, balancing computational efficiency and practicality.

[0084] In one example, an image of a silicone bucket region is uniformly divided according to a preset segmentation rule, resulting in 20 local sub-images of equal area, each with an area of ​​60 pixels (total area = 20 × 60 = 1200 pixels). Through feature comparison, 12 sub-images are determined to be "discolored," with the sum of the discolored areas = 12 × 60 = 720 pixels, and the degree of discoloration = 720 ÷ 1200 = 60%.

[0085] The aforementioned method for detecting discoloration of silicone barrels first filters images containing silicone barrels through target detection and then crops the silicone barrel area to remove background interference, ensuring that subsequent analysis focuses on the silicone barrel area. The core discoloration of silicone in substation breather is "from top to bottom" (moisture and impurities first contact the upper silicone and gradually penetrate downwards). Horizontal segmentation can accurately match this non-uniform distribution feature, avoiding the problem of normal and discolored areas being mixed in the same local sub-image, ensuring that each sub-image can reflect the true state of the corresponding horizontal area. Horizontal segmentation according to preset rules can accurately capture discoloration details at different heights (such as slight discoloration in the upper core area and local discoloration in the lower auxiliary area). Finally, the discolored area spliced ​​from the sub-image states more closely matches the actual discoloration of the silicone barrel. The local sub-image features are compared with a preset feature library to establish an objective and unified judgment standard, and the silicone state is quantified through feature matching. Therefore, it solves the problem of low accuracy in detecting discoloration of breather silicone barrels in existing technologies.

[0086] In one embodiment, the image of the silicone bucket region is segmented horizontally according to a preset segmentation rule to obtain multiple local sub-images, including: dividing the image of the silicone bucket region into an upper core detection area and a lower auxiliary detection area according to a preset segmentation rule; segmenting the upper core detection area horizontally into multiple local sub-images according to a preset fine segmentation rule; and segmenting the lower auxiliary detection area horizontally into multiple local sub-images according to a preset coarse segmentation rule; wherein the segmentation granularity of the fine segmentation rule is smaller than that of the coarse segmentation rule.

[0087] The core design idea of ​​the segmentation rule in this embodiment is to differentiate the segmentation based on the actual color change pattern of the silicone bucket. Since the segmentation granularity of the fine segmentation rule is significantly smaller than that of the coarse segmentation rule, by using different segmentation rules for the core detection area and the auxiliary detection area, the core advantage of improving granularity is retained, and the layered strategy of "fine core area and coarse auxiliary area" takes into account both detection accuracy and computational efficiency.

[0088] This segmentation rule aligns with the actual usage scenario of silicone canisters: discoloration of silicone canisters in substation breathers typically begins from the top (due to moisture and impurities contacting the upper silicone first), making the upper region the core area reflecting discoloration (core detection area), while the lower region has a lower probability and less severe discoloration (auxiliary detection area). Therefore, using "fine segmentation" for the upper region improves the detection granularity of the core area, accurately capturing subtle discoloration features; using "coarse segmentation" for the lower region reduces the number of sub-images without affecting detection accuracy, lowering the computational load for subsequent feature extraction and comparison, thus balancing "accuracy" and "efficiency."

[0089] If a uniform segmentation rule is applied to the entire silicone barrel area, either the overall fine segmentation will lead to computational redundancy, or the overall coarse segmentation will result in inaccurate detection of the core area. This layered segmentation rule, however, specifically addresses the fine-grained detection needs of the core area while avoiding invalid calculations in auxiliary areas, perfectly meeting the actual needs of substation on-site testing.

[0090] In one example, the cropped silicone bucket region image is 400 pixels high. According to a preset segmentation rule, the upper 300 pixels are the core detection area, and the lower 100 pixels are the auxiliary detection area. The core detection area is segmented using a fine segmentation rule (20 pixels / segment), resulting in 15 local sub-images along the horizontal direction; the auxiliary detection area is segmented using a coarse segmentation rule (50 pixels / segment), resulting in 2 local sub-images along the horizontal direction, for a total of 17 local sub-images.

[0091] In another example, the image height of a small silicone bucket region is 250 pixels, and the division rule is 180 pixels at the top (core detection area) and 70 pixels at the bottom (auxiliary detection area). The core detection area is divided into 12 local sub-images according to the fine segmentation rule (15 pixels / segment), and the auxiliary detection area is divided into 2 local sub-images according to the coarse segmentation rule (35 pixels / segment).

[0092] In another example, a silicone bucket showed slight discoloration at its lower part. The area was divided into a 250-pixel upper section (core detection area) and a 150-pixel lower section (auxiliary detection area). The core detection area was divided into 20 local sub-images using a fine segmentation rule (12.5 pixels / segment), while the auxiliary detection area was divided into 5 local sub-images using a relatively fine coarse segmentation rule (30 pixels / segment). This approach ensured the detection accuracy of the core detection area while also meeting the discoloration detection requirements of the auxiliary detection area.

[0093] In one embodiment, the method further includes: calculating an environmental interference coefficient based on the silicone bucket region image; if the environmental interference coefficient is greater than a threshold, reducing the granularity of the fine segmentation rule and the coarse segmentation rule; if the environmental interference coefficient is not greater than the threshold, segmenting the silicone bucket region image according to the conventional granularity of the fine segmentation rule and the coarse segmentation rule.

[0094] This embodiment is a dynamic optimization of the hierarchical segmentation rules. The core logic is to adaptively adjust the segmentation granularity based on environmental interference, so as to ensure detection stability by flexibly adapting to environmental changes and avoiding detection deviations in complex environments with a fixed segmentation granularity.

[0095] The environmental interference coefficient is a quantitative assessment of the image quality of the silicone bucket area (a comprehensive score of interference such as excessively strong / weak lighting, dust obstruction, and image blurring). The threshold is a pre-set "acceptable threshold value for interference". When the interference coefficient is greater than the threshold, the color and texture features of the silicone in the image are easily masked by interference. Reducing the segmentation granularity (i.e., reducing the size and increasing the number of sub-images) can improve the fineness of local features, making it easier for the ResNet50 network to capture the color change details masked by interference and avoid feature extraction distortion caused by interference. When the interference coefficient is not greater than the threshold, the image quality is good and the features are clear. Features can be accurately extracted by using the conventional segmentation granularity, while avoiding computational redundancy caused by excessive segmentation, thus balancing detection accuracy and efficiency.

[0096] Substation environments are complex (e.g., strong sunlight on sunny days, weak sunlight on cloudy days, dust accumulation, etc.), and fixed segmentation granularity cannot adapt to different interference scenarios: when interference is high, conventional segmentation granularity may miss color-changing features, while when interference is low, overly fine granularity will increase computational costs. This dynamic adjustment rule links the segmentation granularity to the environmental interference coefficient, making the segmentation strategy more closely match the real-time image quality, further improving the environmental adaptability and result reliability of the detection method.

[0097] In one example, a substation had a lot of dust, and the image of the silicone barrel area showed obvious dust occlusion, with a calculated environmental interference coefficient of 0.85 (threshold set to 0.6). Because the interference coefficient exceeded the threshold, the segmentation granularity was reduced: the core detection area's regular segmentation granularity of 20 pixels / segment was adjusted to 10 pixels / segment; the auxiliary detection area's regular segmentation granularity of 50 pixels / segment was adjusted to 25 pixels / segment. The original 400-pixel-high silicone barrel (300-pixel core area + 100-pixel auxiliary area) was now divided into 30 local sub-images in the core detection area and 4 sub-images in the auxiliary detection area, for a total of 34 sub-images.

[0098] In another example, the image of the silicone bucket area was acquired under cloudy conditions. The lighting was uniform and there was no obvious occlusion. The calculated environmental interference coefficient was 0.42 (less than the threshold of 0.6). The standard segmentation granularity was applied: 20 pixels / segment for the core detection area and 50 pixels / segment for the auxiliary detection area. The 400-pixel-high silicone bucket was divided into 15+2=17 local sub-images, which controlled the computational load while ensuring accuracy.

[0099] In another example, the image acquired at dusk had slight backlight interference, with an environmental interference coefficient of 0.65 (slightly higher than the threshold of 0.6). The granularity of the segmentation was appropriately reduced: the core detection area was adjusted from 20 pixels / segment to 15 pixels / segment, and the auxiliary detection area from 50 pixels / segment to 30 pixels / segment. The 300-pixel core detection area was divided into 20 segments, and the 100-pixel auxiliary detection area was divided into 3 segments (the remaining 10 pixels were merged into the last sub-image), which both accommodated the interference situation and avoided over-segmentation.

[0100] It should be noted that the core of calculating the environmental interference coefficient is to quantify the degree of interference affecting feature extraction in the image of the silicone bucket area. Considering the characteristics of the substation site environment (lighting, dust, occlusion, blurring, etc.), it can generally be calculated using the method of "weighted summation of multi-dimensional interference indicators". That is, the environmental interference coefficient = Σ (quantized value of each interference dimension × corresponding weight), where the coefficient ranges from [0,1], where 0 represents no interference and 1 represents extreme interference.

[0101] In one embodiment, feature extraction is performed on each local sub-image, and the extracted features are compared with a preset feature library to obtain the comparison results of each local sub-image. This includes: extracting features of the local sub-image based on a preset feature extraction network; calculating the similarity scores of the features with normal silicone features and color-changing silicone features in the feature library based on a similarity algorithm to obtain normal similarity and color-changing similarity; selecting the highest value of a preset number from the normal similarity and color-changing similarity, and calculating the normal average similarity and color-changing average similarity respectively; if the normal average similarity is greater than the color-changing average similarity, the comparison result of the local sub-image is normal; if the normal average similarity is not greater than the color-changing average similarity, the comparison result of the local sub-image is color-changing.

[0102] The core of this embodiment is to improve the accuracy and stability of local subgraph state determination through a closed-loop process of "feature extraction - similarity calculation - mean comparison", and solve the problem that single similarity comparison is easily interfered with by abnormal features.

[0103] In one example, the ResNet50 preset feature extraction network is first used to accurately capture high-dimensional features such as the color and texture of silicone in the local sub-image, providing highly discriminative feature vectors for subsequent comparisons. Second, the cosine similarity algorithm is used to calculate the similarity scores with the "normal / discoloration" features in the feature library. Then, the highest preset number of values ​​are selected and the average similarity is calculated to eliminate individual high / low scores caused by abnormal features in the feature library (such as annotation errors or extreme scene features), reducing the impact of random errors on the judgment results. Finally, the state of the local sub-image is clarified by directly comparing the average similarity of the two types.

[0104] A single similarity score is easily affected by individual abnormal data in the feature library (such as a normal feature having a high similarity to a color-changing sub-image due to labeling errors). However, taking the average of the "highest value of the preset number" can weaken this interference and make the comparison results more accurate.

[0105] In one example, after features were extracted from a local sub-image using ResNet50, 100 similarities (range 0.3-0.95) were calculated with normal silicone features in the feature library, and 100 similarities (range 0.2-0.8) were calculated with color-changing silicone features. With a preset selection of 10, the 10 highest-scoring normal similarities were selected, resulting in an average similarity of 0.92. The 10 highest-scoring color-changing similarities were selected, resulting in an average similarity of 0.78. Since 0.92 > 0.78, the local sub-image was determined to be "normal".

[0106] In one embodiment, the preset feature library is a multi-scene feature library, and its construction process includes: collecting respirator images in different scenarios as sample images; filtering the sample images to obtain sample images containing silicone barrels; cropping the sample images containing silicone barrels to obtain sample silicone barrel images; segmenting the sample silicone barrel images to obtain sample local sub-images; labeling the sample local sub-images as normal samples or discolored samples according to the actual state of the silicone; extracting the local sub-image features of normal samples or discolored samples through a preset feature extraction network, classifying and storing them according to different scenarios to obtain a multi-scene feature library; extracting features from each local sub-image, and comparing the extracted features with the preset feature library, including: determining the target scene feature library matching the current substation from the multi-scene feature library, and comparing the extracted features with the target scene feature library.

[0107] The core of this embodiment is to build a universal feature library that adapts to multiple scenarios, address the core needs of silicone color differences in different scenarios and improve detection universality. Through the logic of "scenario classification construction + targeted matching", it avoids the drawback of existing technologies that require repeated model training.

[0108] The feature library construction process forms a complete closed loop: ① Collect samples from different scenarios (covering different substations, times, weather, and silicone rubber types) to ensure the generalization of the samples and lay the foundation for multi-scenario adaptation; ② Through screening, cropping, and segmentation steps, focus on local sub-images of the core area of ​​the silicone rubber barrel, consistent with the sub-image segmentation rules during actual detection (ensuring feature dimension matching); ③ Manually label the sample status (normal / discolored) to provide accurate category labels for the feature library; ④ Use a feature extraction network (generally a ResNet50 feature extraction network) to extract high-dimensional features and classify and store them according to scenarios, forming a multi-scenario feature library. The core advantage of this embodiment is "distinguishing features according to scenarios," which solves the problem of color differences between normal / discolored silicone rubber in different scenarios.

[0109] In actual testing, the target scenario of the current substation is first matched (such as a 35kV substation, sunny day scenario), and then the feature library of the corresponding scenario is called for comparison, which avoids the judgment error caused by cross-scenario feature mismatch.

[0110] Different substations use different brands of silicone sealant, have varying service lives, and are exposed to different environmental conditions, resulting in variations in the color and texture of normal / color-changing silicone sealant (e.g., normal silicone sealant at station A is blue, while at station B it is light blue). A single feature library cannot adapt to all scenarios. A multi-scenario feature library, however, stores features by categorizing them by scenario, allowing each site to access a feature library tailored to its specific situation. This ensures accurate comparisons and achieves the universal effect of "one algorithm adapting to multiple scenarios," addressing the limitations of existing technologies in terms of versatility.

[0111] In one example, the construction of a multi-scene feature library includes:

[0112] The first step was to collect images: the scenes collected covered 3 types of substations (220kV, 110kV, 35kV), 4 types of weather (sunny, cloudy, rainy, and nighttime supplemental lighting), and 2 types of silicone (Type A and Type B), for a total of 1200 images of respirator samples.

[0113] The second step is to filter and crop: 950 sample images containing complete silicone buckets are selected and cropped to obtain 950 sample silicone bucket images;

[0114] The third step is segmentation and labeling: the samples are segmented according to the segmentation rules during the detection, resulting in 18,200 sample sub-images. Among them, 12,000 are manually labeled as "normal samples" and 6,200 are labeled as "color-changing samples".

[0115] The fourth step is feature extraction and storage: Features of all sample sub-images are extracted using ResNet50 and stored according to the categories of "substation type-weather-silicone type" (such as "220kV-sunny day-type A" and "35kV-night supplementary lighting-type B") to form a multi-scene feature library.

[0116] In one embodiment, after calculating the degree of discoloration of the silicone canister in the respirator, the process includes: comparing the degree of discoloration with a preset warning threshold to obtain the status level of the silicone canister; and generating a status report and a warning signal based on the status level.

[0117] This embodiment is the core link in the practical application of the entire detection process. Its core value is to transform the quantified degree of color change into status conclusions and action instructions that can directly guide operation and maintenance work.

[0118] The core objective of this embodiment is to serve the maintenance of respirators in power grid scenarios. Simply outputting the degree of color change cannot directly guide the work. This step, through a closed loop of "threshold comparison - level determination - information output," transforms the detection technology into practical maintenance capabilities: it avoids the errors of subjective judgment by maintenance personnel, and enables "differentiated response" through graded early warning (e.g., slight color change does not require immediate treatment, severe color change requires emergency replacement), reducing maintenance costs and ensuring the safety of power grid equipment.

[0119] In one example, the preset warning thresholds are: Level 1 threshold (normal) ≤30%, Level 2 threshold (warning) 30%-70%, and Level 3 threshold (emergency) >70%. The detection result is that the silicone barrel discoloration degree is 45%, which is determined to be a "Level 2 warning" state after comparing with the threshold. The output status report includes information such as "discoloration degree 45% and discoloration of the upper 15%-30% area of ​​the core detection area". The warning signal is a pop-up prompt from the substation operation and maintenance system + SMS notification, reminding that "silicone inspection or replacement should be arranged within 72 hours".

[0120] In one embodiment, a device for detecting color change in a silicone bucket is provided. For example... Figure 2 As shown, the silicone barrel color change detection device includes an acquisition module 21, a screening module 22, a segmentation module 23, a comparison module 24, and a calculation module 25. Detailed descriptions of each functional module are as follows:

[0121] Acquisition module 21 is used to acquire images of the respirator in the substation;

[0122] The filtering module 22 is used to perform target detection on the respirator image, filter out the respirator image containing the silicone barrel as the target image, and extract the silicone barrel area image from the target image.

[0123] The segmentation module 23 is used to segment the image of the silicone bucket region along the horizontal direction according to a preset segmentation rule to obtain multiple local sub-images;

[0124] The comparison module 24 is used to extract features from each local sub-image, compare the extracted features with a preset feature library, and obtain the comparison results of each local sub-image.

[0125] The calculation module 25 is used to calculate the degree of discoloration of the silicone barrel in the respirator based on the comparison results of each local sub-image.

[0126] This invention also provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the aforementioned silicone bucket color-changing detection method; to avoid repetition, this will not be described again here. Alternatively, the electronic device can implement the functions of each module in this embodiment of the silicone bucket color-changing detection device; this will also not be described again here.

[0127] This invention also provides a readable storage medium storing a program. When executed by a processor, the program implements the aforementioned silicone bucket color change detection method; to avoid repetition, this will not be described again here. Alternatively, when executed by a processor, the program implements the functions of each module in this embodiment of the silicone bucket color change detection device; this will also not be described again here.

[0128] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

[0129] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0130] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for detecting color change in a silicone bucket, characterized in that, include: Acquire images of the respirator in the substation; Target detection is performed on the respirator image to filter out the respirator image containing the silicone barrel as the target image, and the silicone barrel area image is cropped from the target image; The image of the silicone bucket region is divided horizontally according to a preset segmentation rule to obtain multiple local sub-images; Feature extraction is performed on each of the local sub-images, and the extracted features are compared with a preset feature library to obtain the comparison results of each local sub-image. Based on the comparison results of the various local sub-images, the degree of discoloration of the silicone barrel in the respirator is calculated; The step of extracting features from each of the local sub-images and comparing the extracted features with a preset feature library to obtain the comparison results of each local sub-image includes: Based on a preset feature extraction network, the features of the local sub-image are extracted; Based on the similarity algorithm, the similarity scores between the feature and the normal silicone feature and the color-changing silicone feature in the feature library are calculated respectively to obtain the normal similarity and the color-changing similarity. The highest values ​​of the preset number are selected from both the normal similarity and the color-changing similarity, and the average similarity of the normal and color-changing samples are calculated respectively. If the normal average similarity is greater than the color-changing average similarity, the comparison result of the local sub-image is normal; if the normal average similarity is not greater than the color-changing average similarity, the comparison result of the local sub-image is color-changing. The image of the silicone bucket region is segmented horizontally according to a preset segmentation rule to obtain multiple local sub-images, including: The image of the silicone bucket area is divided into an upper core detection area and a lower auxiliary detection area according to a preset division rule; According to the preset fine segmentation rules, the upper core detection area is divided into multiple local sub-images along the horizontal direction; According to the preset coarse segmentation rules, the lower auxiliary detection area is divided into multiple local sub-images along the horizontal direction; The fine-grained segmentation rule has a smaller segmentation granularity than the coarse-grained segmentation rule; Calculate the environmental interference coefficient based on the image of the silicone bucket area; If the environmental interference coefficient is greater than the threshold, then the granularity of the fine segmentation rule and the coarse segmentation rule is reduced. If the environmental interference coefficient is not greater than the threshold, the silicone bucket region image is segmented according to the conventional segmentation granularity of fine segmentation rules and coarse segmentation rules.

2. The method for detecting discoloration of a silicone bucket according to claim 1, characterized in that, The calculation of the degree of discoloration of the silicone tubing in the respirator based on the comparison results of each local sub-image includes: Obtain the comparison results of each local sub-image, the comparison results including normal or discolored; Obtain the area of ​​each local sub-image that changes color in the comparison result; The ratio of the sum of the areas of the discolored local sub-images to the sum of the areas of all local sub-images is calculated, and this ratio is the degree of discoloration of the silicone barrel in the respirator.

3. The method for detecting discoloration of a silicone bucket according to claim 1, characterized in that, The preset feature library is a multi-scene feature library, and its construction process includes: Images of respirators in different scenarios were collected as sample images; The sample images are filtered to obtain sample images containing silicone buckets; The sample image containing the silicone bucket is obtained by cropping the sample image. The sample silicone bucket image is segmented to obtain a sample local sub-image; The local sub-images of the samples are labeled as normal samples or discolored samples according to the actual state of the silicone. By using a preset feature extraction network, local sub-image features of normal samples or color-changing samples are extracted, and classified and stored according to different scenarios to obtain the multi-scenario feature library; The step of extracting features from each of the local sub-images and comparing the extracted features with a preset feature library includes: A target scene feature library matching the current substation is determined from a multi-scene feature library, and the extracted features are compared with the target scene feature library.

4. The method for detecting discoloration of a silicone bucket according to claim 1, characterized in that, The calculation of the degree of discoloration of the silicone tubing in the respirator includes: By comparing the degree of discoloration with a preset warning threshold, the state level of the silicone bucket is obtained; Status reports and warning signals are generated based on the status level.

5. A device for detecting color change in a silicone bucket, characterized in that, include: The acquisition module is used to acquire images of the respirator in the substation; The filtering module is used to perform target detection on the respirator image, filter out the respirator image containing the silicone barrel as the target image, and extract the silicone barrel area image from the target image; The segmentation module is used to segment the image of the silicone bucket region along the horizontal direction according to a preset segmentation rule to obtain multiple local sub-images; The comparison module is used to extract features from each of the local sub-images, compare the extracted features with a preset feature library, and obtain the comparison results of each local sub-image. The calculation module is used to calculate the degree of discoloration of the silicone barrel in the respirator based on the comparison results of each local sub-image; The step of extracting features from each of the local sub-images and comparing the extracted features with a preset feature library to obtain the comparison results of each local sub-image includes: Based on a preset feature extraction network, the features of the local sub-image are extracted; Based on the similarity algorithm, the similarity scores between the feature and the normal silicone feature and the color-changing silicone feature in the feature library are calculated respectively to obtain the normal similarity and the color-changing similarity. The highest values ​​of the preset number are selected from both the normal similarity and the color-changing similarity, and the average similarity of the normal and color-changing samples are calculated respectively. If the normal average similarity is greater than the color-changing average similarity, the comparison result of the local sub-image is normal; if the normal average similarity is not greater than the color-changing average similarity, the comparison result of the local sub-image is color-changing. The image of the silicone bucket region is segmented horizontally according to a preset segmentation rule to obtain multiple local sub-images, including: The image of the silicone bucket area is divided into an upper core detection area and a lower auxiliary detection area according to a preset division rule; According to the preset fine segmentation rules, the upper core detection area is divided into multiple local sub-images along the horizontal direction; According to the preset coarse segmentation rules, the lower auxiliary detection area is divided into multiple local sub-images along the horizontal direction; The fine-grained segmentation rule has a smaller segmentation granularity than the coarse-grained segmentation rule; Calculate the environmental interference coefficient based on the image of the silicone bucket area; If the environmental interference coefficient is greater than the threshold, then the granularity of the fine segmentation rule and the coarse segmentation rule is reduced. If the environmental interference coefficient is not greater than the threshold, the silicone bucket region image is segmented according to the conventional segmentation granularity of fine segmentation rules and coarse segmentation rules.

6. A computer 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, it implements the steps of the silicone bucket color change detection method as described in any one of claims 1 to 4.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the silicone bucket color change detection method as described in any one of claims 1 to 4.

Images