Machine vision-based communication room indicator light state intelligent inspection and early warning method

Abstract

The application discloses a communication machine room indicating lamp state intelligent inspection and early warning method based on machine vision, and relates to the technical field of machine vision; the method comprises the following steps: collecting original images of machine room indicating lamps and preprocessing, extracting core bright area images, fusing membership, non-membership and hesitancy to construct fuzzy features, combining the core bright area and the fuzzy features to calculate pixel attribution probability, performing weighted clustering iteration optimization according to the probability, obtaining a membership matrix and an optimal clustering center, segmenting the final image according to the membership matrix and the optimal clustering center, inputting a state analysis model to identify abnormalities and trigger early warning; the method preprocesses the images of the machine room indicating lamps and extracts the core bright area, fuses multi-dimensional fuzzy features to calculate pixel attribution probability, and optimizes the membership matrix and the clustering center by means of a clustering algorithm, so that the method can effectively resist the interference of machine room illumination, sundries and environmental interference, strengthen the feature representation effect, greatly improve the indicating lamp image segmentation and state recognition precision, and guarantee the safe operation and maintenance of the machine room equipment and the inspection efficiency.

Description

Technical Field

[0001] This invention belongs to the field of machine vision technology, specifically relating to a machine vision-based intelligent inspection and early warning method for indicator light status in communication equipment rooms. Background Technology

[0002] Indicator lights in communication equipment rooms serve as status indicators for various communication devices, power supplies, and supporting facilities. They can intuitively reflect the real-time operating status of equipment, fault alarms, and connectivity status. With densely packed and numerous devices in equipment rooms, abnormal indicator lights often conceal potential equipment faults. Existing manual inspections are prone to omissions, false positives, and low efficiency. Intelligent status inspection of indicator lights can accurately capture abnormal changes in light status, promptly identify potential equipment problems, reduce manual maintenance costs, avoid human oversights, ensure the stable and reliable operation of communication equipment, and effectively improve the level of intelligent operation and maintenance in equipment rooms and the overall operational security of communication networks.

[0003] In complex operation and maintenance scenarios of communication equipment rooms, the metal cabinets of equipment racks are prone to producing bright reflections and scattered isolated bright spots. Their grayscale characteristics are highly similar to the bright areas of indicator lights. At the same time, indicator lights exhibit halo diffusion, causing blurred grayscale transitions at the edges. Existing indicator light recognition technologies mostly rely on fixed grayscale threshold segmentation, traditional fuzzy C-means clustering, or single target detection methods. They only utilize the inherent grayscale information of pixels for analysis, which cannot effectively model the fuzzy uncertainty of pixel grayscale and lacks constraints on the spatial neighborhood distribution of pixels. This easily leads to misjudging cabinet reflections and invalid noise as indicator light targets, resulting in rough segmentation boundaries and severe false target interference. Moreover, existing methods for indicator light status discrimination mostly rely on manually set fixed thresholds, which are difficult to adapt to the changes in pixel grayscale drift caused by fluctuations in equipment room lighting and long-term equipment aging. Furthermore, they fail to achieve adaptive threshold updates through cluster optimal center priors, resulting in low target segmentation accuracy and insufficient reliability of status recognition for equipment room indicator lights. This makes it difficult to meet the needs of high-precision, interference-resistant automatic detection and status identification of indicator lights in complex equipment room environments. Summary of the Invention

[0004] The purpose of this invention is to solve the problems of low accuracy in indicator light status detection caused by the confusion between the metal reflection of the cabinet, isolated bright spots and the grayscale features of the indicator lights, and the blurred grayscale of the indicator light halo edges, which leads to pixel grayscale drift in indicator light image recognition and failure to achieve adaptive threshold update through clustering optimal center prior. Therefore, this invention proposes a machine vision-based intelligent inspection and early warning method for indicator light status in communication equipment rooms.

[0005] The present invention proposes a machine vision-based intelligent inspection and early warning method for indicator light status in communication equipment rooms, the method comprising: The original images of the indicator lights in the communication equipment room are acquired, and the original images are preprocessed to obtain an initial image; Extract the core bright area image of the computer room indicator light from the initial image, and perform feature extraction based on the initial image to obtain fuzzy features; the fuzzy features include optimized membership degree, non-membership degree, and hesitation degree; The probability of each pixel belonging to an indicator light is calculated based on the core bright area image and blur features. The final clustering membership matrix and optimal cluster center of the indicator light are obtained by a preset clustering algorithm based on the membership probability. The final image is obtained by segmentation based on the final clustering membership matrix and the optimal cluster center. The final image is input into the state analysis model to obtain the analysis results of the indicator light status, and an early warning process is performed based on the analysis results. The state analysis model is based on an improvement of YOLOv8, and the improvement includes: replacing all C2f modules in the backbone network and neck network of YOLOv8 with target C2f modules. The target C2f modules are used to improve the extraction capability of key features of the indicator light status.

[0006] Optionally, the original image is preprocessed to obtain an initial image, including: The original image is converted into a grayscale image, and the brightness value and spatial coordinates of the target pixel are extracted; the target pixel is any pixel in the grayscale image. The membership and non-membership of the target pixel are initialized based on the brightness value of the target pixel; The brightness value, spatial coordinates, membership degree and non-membership degree of the target pixel are bound and fused to generate an information-containing target pixel. The information-containing target pixel is then filled into the corresponding position in the original image to obtain the initial image.

[0007] Optionally, the core bright area image of the server room indicator lights in the initial image is extracted, including: Perform brightness saliency detection processing on the initial image to output a brightness saliency image of the indicator light; The first pixel is obtained by filtering each row of pixels in the bright image by using a brightness threshold, and the area where the first pixel is located is used as the first candidate indicator area. The first brightness data of the pixels within the first candidate indicator light area is statistically analyzed, and the indicator light row length data and center horizontal coordinate data of the first candidate indicator light area are calculated based on the first brightness data; the first brightness data includes the sum of the brightness of the pixels within the first candidate indicator light area, the maximum brightness value, and the total number of pixels; The core pixel of the indicator light in each row is determined based on the row length data and center horizontal coordinate data of the candidate indicator light area in each row. The core pixels of each row of indicator lights are then stitched together to obtain the indicator light row image. The second pixel is obtained by filtering each column of pixels in the bright image by using a brightness threshold, and the area where the second pixel is located is used as the second candidate indicator area. The second brightness data of the pixels within the second candidate indicator area is statistically analyzed, and the indicator length data and center vertical coordinate data of the second candidate indicator area are calculated based on the second brightness data; the second brightness data includes the sum of the brightness of the pixels within the second candidate indicator area, the maximum brightness value, and the total number of pixels; The core pixel of the indicator light in each column is determined based on the column length data and center vertical coordinate data of the candidate indicator light area of ​​each column of pixels. The core pixels of each column of indicator lights are then stitched together to obtain the indicator light column image. The row image and column image of the indicator lights are fused together to obtain the core bright area image of the computer room indicator lights.

[0008] Optionally, feature extraction is performed on the initial image to obtain blurred features, including: The brightness threshold is calculated, and the brightness boundary is used to segment each pixel in the initial image to obtain the optimized membership degree of each pixel belonging to the indicator light. The brightness boundary includes a low brightness boundary and a high brightness boundary. The non-membership degree and hesitation degree of each pixel are calculated using a preset algorithm based on the optimized membership degree of each pixel.

[0009] Optionally, the probability of each pixel belonging to an indicator light is calculated based on the core bright area image and blur features, including: Based on the initial image, all pixels in the core region of the core bright area image are selected and a valid pixel set is formed. Taking any pixel in the initial image as the center, select pixels within an 8-neighborhood, count the number of pixels within the neighborhood that belong to the effective pixel set, and calculate the neighborhood weight factor based on the adjustment coefficient. The probability of each pixel belonging to an indicator light is calculated based on the fuzzy features and the neighborhood weight factor.

[0010] Optionally, a preset clustering algorithm is used to obtain the final clustering membership matrix and optimal cluster centers of the indicator lights based on the membership probability, including: Step 1: Randomly select a preset number of pixel grayscale features in the initial image as the initial cluster centers, and initialize the initial cluster membership. Step 2: Construct the target constraint function based on the membership probability; the target constraint function is used to find the optimal cluster center and cluster assignment membership degree to achieve accurate fuzzy clustering of indicator light targets and complex backgrounds; Step 3: Under the constraint of minimizing the objective constraint function, update the initial cluster membership degree pixel by pixel to obtain the updated cluster membership degree matrix; Step 4: Update the initial cluster centers based on the cluster membership matrix and the membership probability to obtain the updated cluster centers; Step 5: Set the iteration termination condition based on the previous clustering membership matrix and the current clustering membership matrix. If the iteration termination condition is met or the maximum number of iterations is reached, stop the iteration and output the final clustering membership matrix and the optimal cluster center. Otherwise, execute steps 3 and 4 to enter the next round of iteration.

[0011] Optionally, the final image is obtained by segmentation based on the final clustering membership matrix and the optimal cluster centers, including: Based on the optimal cluster center, the indicator light target cluster category and the computer room background cluster category are divided. Combining the membership degree values ​​of each pixel in the final cluster membership degree matrix, the membership degree judgment threshold is set. Pixels whose cluster membership degree is greater than the membership degree determination threshold and belong to the target cluster category of the indicator light are marked as foreground pixels of the indicator light, and the remaining pixels are marked as background pixels of the computer room. The marked pixels are subjected to connected component noise reduction to obtain a complete and continuous connected component of the indicator light. The pixels in the connected component of the indicator light are then reconstructed and fused with the background pixels at the pixel level to generate the final image.

[0012] Optionally, the state analysis model is an improvement based on YOLOv8, and the improvements include: All C2f modules in the backbone and neck networks of YOLOv8 are replaced with target C2f modules; the target C2f modules are used to enhance the flow and reuse of gradient information in the complex environment of the computer room and improve the ability to extract key features of indicator light status.

[0013] Optionally, the target C2f module is an improvement based on the C2f module, including: Replace the DarknetBottleneck module in the C2f module with the target Darknet module; The working principle of the target Darknet module is as follows: Obtain the first input feature tensor, split the input feature tensor in half according to the channel dimension to obtain the first feature tensor and the second feature tensor, input the first feature tensor into the CBS module and two concatenated target modules in sequence to obtain the third feature tensor, and input the second feature tensor into the CBS module to obtain the fourth feature tensor; The third and fourth feature tensors are concatenated and then input into the CBS module to obtain the fifth feature tensor. The working principle of the target module is as follows: Obtain the second input feature tensor, and input the second input feature tensor into two parallel Conv modules and one BN module respectively to obtain the first convolutional feature tensor, the second convolutional feature tensor and the third convolutional feature tensor; The first and second convolutional feature tensors are input into two BN modules respectively, and the fourth and fifth convolutional feature tensors are output. The 3rd, 4th, and 5th convolutional feature tensors are concatenated and then sequentially input into the Conv, BN, and CBS modules to output the 6th convolutional feature tensor.

[0014] Optionally, early warning processing is performed based on the analysis results, including: Obtain the indicator light status category output by the status analysis model; the status category includes normal status and at least one abnormal status, the abnormal status includes indicator light not lit, indicator light always lit error, indicator light flashing frequency abnormal, indicator light color error; When the status category is normal, continue to execute the next round of image acquisition and inspection; When the state category is an abnormal state, the abnormality type and confidence score are extracted from the analysis results. A preset warning level is matched based on the anomaly type and confidence level value. The warning levels include Level 1 warning, Level 2 warning, and Level 3 warning. Execute the corresponding warning operation according to the warning level; the warning operation includes at least one of the following: generate an abnormal log and store it in the database, highlight the abnormal indicator light position on the monitoring interface and display the status details, send an alarm message to the operation and maintenance personnel terminal, and trigger the on-site audible and visual alarm; If the same indicator light repeatedly triggers the same level of warning within the preset continuous inspection cycle, the warning level will be automatically upgraded and higher-level maintenance personnel will be notified.

[0015] The beneficial effects of this invention are: This invention proposes a machine vision-based intelligent inspection and early warning method for indicator lights in communication equipment rooms. It preprocesses the original images of the indicator lights, extracts the core bright areas, and integrates fuzzy features with optimized membership, non-membership, and hesitation to accurately calculate pixel ownership probabilities. Furthermore, it optimizes the cluster membership matrix and cluster centers using a clustering algorithm, improving image segmentation accuracy, effectively avoiding interference from lighting conditions and environmental clutter, enhancing the indicator light feature representation capabilities, and significantly improving the accuracy of indicator light region segmentation and status recognition. This ensures the operational safety and inspection efficiency of communication equipment room equipment. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 A flowchart illustrating the intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision, provided in an embodiment of the present invention. Figure 2 A schematic diagram of the network structure of the state analysis model provided in this embodiment of the invention; Figure 3 An architecture diagram of the target Darknet module provided in an embodiment of the present invention; Figure 4 This is an architecture diagram of the target module provided in an embodiment of the present invention. Detailed Implementation

[0018] 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 embodiments of the present invention, and not all embodiments. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and B can represent: A alone, A and B simultaneously, and B alone. Furthermore, descriptions involving "first," "second," etc., in this invention are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" can explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0019] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] This invention provides a machine vision-based intelligent inspection and early warning method for indicator light status in communication equipment rooms. See also... Figure 1 , Figure 1 A flowchart illustrating a machine vision-based intelligent inspection and early warning method for indicator light status in a communication equipment room, provided in an embodiment of the present invention. The method includes the following steps: S1, acquire the original image of the indicator lights in the communication equipment room, and preprocess the original image to obtain the initial image; S2, extract the core bright area image of the computer room indicator light in the initial image, and extract the fuzzy features based on the initial image; S3, calculate the indicator light attribution probability of each pixel based on the core bright area image and blur features; S4. Based on the membership probability, a preset clustering algorithm is used to obtain the final clustering membership matrix and the optimal cluster center of the indicator light. The final image is obtained by segmentation based on the final clustering membership matrix and the optimal cluster center. S5. Input the final image into the state analysis model to obtain the analysis results of the indicator light status, and perform early warning processing based on the analysis results.

[0021] Fuzzy features include optimized membership, non-membership, and hesitation.

[0022] The state analysis model is based on an improvement of YOLOv8. The improvements include replacing all C2f modules in the backbone and neck network of YOLOv8 with target C2f modules. The target C2f modules are used to improve the ability to extract key features of indicator light states.

[0023] The intelligent inspection and early warning method for indicator lights in communication equipment rooms based on machine vision provided in this invention preprocesses the original images of the indicator lights, extracts the core bright areas, and integrates fuzzy features including optimized membership, non-membership, and hesitation to accurately calculate pixel ownership probabilities. It also optimizes the clustering membership matrix and cluster centers using a clustering algorithm to improve image segmentation accuracy, effectively avoids interference from lighting and environmental clutter in the equipment room, strengthens the indicator light feature representation capabilities, and significantly improves the accuracy of indicator light area segmentation and status recognition, ensuring the operational safety and inspection efficiency of communication equipment room equipment.

[0024] In one implementation, membership degree represents the degree to which a pixel belongs to an indicator light, non-membership degree represents the degree to which a pixel does not belong to an indicator light, and hesitation degree represents the degree of uncertainty (e.g., unclear judgment due to halo, blur, or interference).

[0025] In one embodiment, preprocessing the original image to obtain an initial image includes: The original image is converted into a grayscale image, and the brightness value and spatial coordinates of the target pixel are extracted; the target pixel is any pixel in the grayscale image. Initialize the membership and non-membership of the target pixel based on its brightness value; The brightness value, spatial coordinates, membership degree and non-membership degree of the target pixel are bound and fused to generate an information-containing target pixel. The information-containing target pixel is then filled into the corresponding position in the original image to obtain the initial image.

[0026] In one implementation, the membership degree of the brightness value of the target pixel is obtained by normalizing the brightness value of the target pixel; The non-membership degree is calculated based on the membership degree, and the process is as follows:

[0027] in, For the j-th pixel, Let represent the initial non-membership degree of the j-th pixel. This represents the initial membership degree of the j-th pixel. The parameter was set to 0.5 in the experiment to be used for adjustment. The closer a pixel is to 1, the less likely it is to be an indicator light.

[0028] In one implementation, the brightness features, two-dimensional spatial coordinates, and membership and non-membership features of each pixel in the visible light real-shot image of the computer room cabinet are fused and bound together, so that it does not rely solely on the limitations of a single brightness feature, thus laying a reliable data foundation for the extraction of the core bright area of ​​the indicator light.

[0029] In one embodiment, extracting the core bright area image of the server room indicator lights from the initial image includes: Perform brightness saliency detection processing on the initial image and output a brightness saliency image of the indicator light; The first pixel is obtained by filtering each row of pixels in the image with significant brightness using a brightness threshold, and the area where the first pixel is located is used as the first candidate indicator area. The first brightness data of the pixels within the first candidate indicator light area is statistically analyzed. Based on the first brightness data, the indicator light row length data and center horizontal coordinate data of the first candidate indicator light area are calculated. The first brightness data includes the total brightness of the pixels within the first candidate indicator light area, the maximum brightness value, and the total number of pixels. The core pixel of the indicator light in each row is determined based on the row length data and center horizontal coordinate data of the candidate indicator light area in each row. The core pixels of each row of indicator lights are then stitched together to obtain the indicator light row image. The second pixel is obtained by filtering each column of pixels in the image with significant brightness by using a brightness threshold, and the area where the second pixel is located is used as the second candidate indicator area. The second brightness data of pixels within the second candidate indicator area is statistically analyzed. Based on the second brightness data, the indicator length data and center vertical coordinate data of the second candidate indicator area are calculated. The second brightness data includes the sum of the brightness of pixels within the second candidate indicator area, the maximum brightness value, and the total number of pixels. The core pixel of the indicator light in each column is determined based on the column length data and center vertical coordinate data of the candidate indicator light area of ​​each column of pixels. The core pixels of each column of indicator lights are then stitched together to obtain the indicator light column image. The core bright area image of the indicator lights in the computer room is obtained by fusing the row image and column image of the indicator lights.

[0030] In one implementation, brightness saliency detection processing is performed on the initial image, including: Pixel-level brightness saliency values ​​are constructed by calculating the brightness difference and contrast between each pixel and its local neighboring pixels in the initial image. The high-brightness indicator areas in the initial image are assigned high saliency values, while low-brightness background areas (such as cabinet panels, cables, dust, etc.) are assigned low saliency values. After background suppression and indicator area enhancement are completed, the output is a saliency map that only highlights the candidate areas of the indicator lights.

[0031] In one implementation, it should be noted that each row of the image with significant brightness has more than one first candidate indicator light area, and each column has more than one second candidate indicator light area. The number of candidate indicator light areas is determined by the selected pixels. For example, if the light-emitting areas of three lights all fall on the horizontal pixel line of the 10th row of the image, then three candidate indicator light areas will be identified.

[0032] In one implementation, the first pixel is obtained by filtering the pixels in each row of the image with significant brightness. The process is as follows: The brightness threshold of the indicator light is calculated from a luminous image using the Otsu algorithm. Pixels with brightness values ​​not less than a brightness threshold are selected from the bright images as the first pixel that meets the criteria.

[0033] In one implementation, the indicator length data and coordinate data of the candidate indicator area are calculated based on the brightness data, including:

[0034]

[0035]

[0036]

[0037] in, This indicates the i-th candidate indicator light area in the current row. Let j be the brightness value of the j-th pixel. This is the brightness threshold. This represents the length data of the i-th candidate indicator area in the current row. This represents the total pixel brightness of the i-th candidate indicator area in the current row. This represents the maximum brightness value of the pixel in the i-th candidate indicator area of ​​the current row; This represents the horizontal coordinate of the center of the i-th candidate indicator area in the current row. The x-coordinate of the j-th pixel This indicates the total number of pixels in the i-th candidate indicator area of ​​the current row; This represents the vertical coordinate of the center of the i-th candidate indicator area in the current row. The ordinate of the j-th pixel.

[0038] In one implementation, the core pixel of the indicator light for each row of pixels is determined based on the row length data and horizontal coordinate data of the candidate indicator light area. Start column = center coordinates - length ÷ 2, end column = center coordinates + length ÷ 2; Based on the start and end columns, determine which pixels in this row belong to the indicator light's core pixels; Pixels falling within the core area are marked as 1, indicating the core area (representing the core of the indicator light). Pixels outside the specified range are marked as 0, representing the background area; For example, in row 55, for an indicator light: the center horizontal coordinate is 130 and the length data is 30, then: the starting column = 130 - 15 = 115, and the ending column = 130 + 15 = 145; The pixels in row 55 and between columns 115 and 145 are the core pixels of the indicator light in that row. If the pixels are not between columns 115 and 145, they represent the background area. The calculation method for the core pixels of each column of pixel indicator lights is the same.

[0039] In one implementation, the row image and column image of the indicator lights are subjected to set union fusion processing to obtain the core bright area data, retaining complete and unmissing core pixels of the indicator lights, and avoiding core omissions caused by a single operation sequence.

[0040] In one implementation, due to the complex background interference such as strong reflections from metal panels, horizontal bright spots, dust noise, and cable shadows in the communication equipment room cabinet, and the lack of fixed geometric shapes, dense arrangement of multiple lights, and blurred halo boundaries of the on-site indicator lights, single threshold segmentation or unidirectional screening is prone to misjudging strip-shaped reflections and scattered bright spots as indicator lights. By performing bidirectional threshold screening and core parameter calculation on the brightness saliency map row by row and column by column, without relying on the prior knowledge of the indicator light shape, and with dual constraint verification from both horizontal and vertical dimensions, interference targets such as horizontal reflective strips and invalid bright noise that cannot be eliminated by unidirectional screening are effectively filtered out. The center coordinates and effective core range of each indicator light are accurately located, and finally, the pure core bright area of ​​the indicator light is extracted, thereby weakening the complex background interference and eliminating false target interference from the source.

[0041] In one embodiment, feature extraction from an initial image yields blurred features, including: The brightness boundary is calculated based on the brightness threshold. Based on the brightness boundary, each pixel in the initial image is segmented and the optimized membership degree of each pixel belonging to the indicator light is calculated. The brightness boundary includes the low brightness boundary and the high brightness boundary. The non-membership degree and hesitation degree of each pixel are calculated using a preset algorithm based on the optimized membership degree of each pixel.

[0042] In one implementation, the brightness boundary is calculated based on a brightness threshold, including: The brightness threshold is normalized to obtain the normalization coefficient. Based on the normalization coefficient and the preset expansion coefficient, the low-brightness boundary and high-brightness boundary that distinguish the background of the server rack are calculated respectively.

[0043]

[0044] in, The normalization coefficient is... and These are the minimum and maximum brightness values ​​of the pixels in the initial image, respectively; k is the preset dilation coefficient; and These represent the low-brightness boundary and the high-brightness boundary, respectively.

[0045] In one implementation, the optimized membership degree of each pixel belonging to the indicator light is calculated by segmenting each pixel in the initial image according to the brightness boundary.

[0046] in, Represents pixels Optimized membership degree of indicators belonging to the communication equipment room.

[0047] In one implementation, the non-membership degree and hesitation degree of each pixel are calculated using a preset algorithm based on the optimized membership degree of each pixel. The calculation formula is as follows:

[0048]

[0049] in, Represents pixels The non-membership degree of indicators that do not belong to the communication equipment room indicator lights. For pixels The degree of hesitation is used to characterize the blurring uncertainty of indicator light halos and metallic reflective edge pixels; The value of the control parameter was set to 0.5 in the experiment.

[0050] In one implementation, the existing single linear blurring method is difficult to adapt to the complex imaging characteristics of communication equipment rooms. Metal reflections from the cabinet, dust noise, and the diffusion of LED halo will cause smooth pixel grayscale transitions and grayscale mixing between the target and the background. Conventional blurring is difficult to accurately depict the halo edge and the uncertainty of the attribution of reflective pixels, which easily leads to target misjudgment and blurred boundary division. By constructing a dedicated fuzzification model by dividing the grayscale range into three segments—the low-brightness background area, the halo transition area, and the high-brightness core area of ​​the indicator light—and solving the membership, non-membership, and hesitation degrees pixel by pixel, the fuzziness degree of different grayscale pixels belonging to the indicator light can be accurately quantified. This effectively distinguishes the core area of ​​the indicator light, the halo transition area, and the cabinet background area, avoiding the shortcomings of the insufficient distinguishing ability of existing linear fuzzification and suppressing feature confusion caused by background grayscale interference such as metallic reflections.

[0051] In one embodiment, the probability of each pixel belonging to an indicator light is calculated based on the core bright area image and blur features, including: Based on the initial image, all pixels in the core region of the core bright area image are selected and formed into a valid pixel set; Taking any pixel in the initial image as the center, select pixels within an 8-neighborhood, count the number of pixels in the neighborhood that belong to the effective pixel set, and calculate the neighborhood weight factor with the adjustment coefficient. The probability of each pixel belonging to an indicator light is calculated based on the fuzzy features and the neighborhood weight factor.

[0052] In one implementation, all pixels in the initial image are traversed, and the pixel positions are matched and compared with the core bright area image. All pixels corresponding to the core region in the core bright area image are selected and defined as the effective pixel set of the indicator light.

[0053] In one implementation, the pixels of the initial image and the core bright area image are filtered by coordinate matching to accurately define the core effective pixel set of the indicator light. This aims to distinguish the real indicator light from non-target pixels such as background and reflective noise in the computer room image, thereby reducing redundant calculations.

[0054] In one implementation, the neighborhood weight factor calculation includes:

[0055] in, Let be the spatial neighborhood weight factor for the j-th pixel, and exp be the natural exponential function. This is the weight adjustment coefficient, used to control the sensitivity of neighborhood association; the experimental value is 0.8. This represents the number of pixels in the neighborhood that belong to the valid pixel set. This represents the number of pixels in the 8-neighborhood.

[0056] In one implementation, spatial structure constraints are introduced by utilizing the characteristics of the continuous distribution of indicator light pixels and the discrete single-point reflection of isolated reflections: real indicator lights with a large number of core pixels in the surrounding area are given higher weights, while isolated pseudo-targets without neighborhood support are given weaker weights. This makes up for the inability of grayscale blur features to distinguish single-point bright interference, distinguishes targets from noise from a spatial dimension, and effectively suppresses interference from isolated reflections and scattered bright spots.

[0057] In one implementation, the probability of each pixel belonging to an indicator light is calculated based on fuzzy features and neighborhood weight factors, including:

[0058] in, Represents pixels The initial probability of belonging; For the initial attribution probability After normalization, the attribution probability is obtained.

[0059] In one implementation, the positive, negative, and fuzzy uncertainties of pixel attribution are fully covered by three states: membership, non-membership, and hesitation. This makes up for the shortcomings of a single membership degree and compensates for the hesitation information by allocating it to the target attribution dimension. The attribution probability is derived by integrating the three types of fuzzy information. By fully analyzing the visual fuzzy features of pixel grayscale, taking into account the differences between the definite region and the fuzzy transition region, and combining spatial neighborhood weights to correct the credibility, the classification ambiguity of reflection, noise, and halo boundaries is weakened, and the overall credibility of the pixel attribution indicator is accurately quantified.

[0060] In one implementation, due to issues such as isolated metallic reflections from the equipment room, scattered bright noise, and blurred pixel grayscale at the edge of the indicator light halo in the images, relying solely on a single visual blur feature cannot simultaneously consider pixel grayscale attributes and spatial location correlation. The lack of neighborhood constraints makes it easy to misjudge false target bright spots as indicator light pixels. Therefore, it is necessary to define an effective pixel set based on the previously extracted core bright area of ​​the indicator light as a spatial prior, calculate the spatial correlation weight by combining the pixel's eight-neighborhood structural features, and then fuse membership, non-membership, and hesitation fuzzy information and introduce uncertainty compensation. After global normalization, a standard induced assignment probability is obtained. This organically combines the pixel grayscale blur characteristics with the spatial neighborhood constraint relationship, weakens the interference of isolated reflections and invalid noise, and accurately quantifies the credibility of each pixel belonging to the core area of ​​the indicator light.

[0061] In one embodiment, a preset clustering algorithm is used to obtain the indicator light clustering membership matrix and the optimal cluster centers based on the membership probability, including: Step 1: Randomly select a preset number of pixel grayscale features in the initial image as the initial cluster centers, and initialize the initial cluster membership. Step 2: Construct the target constraint function based on the membership probability; the target constraint function is used to find the optimal cluster center and assign the membership degree of the cluster, so as to achieve accurate fuzzy clustering of indicator light targets and complex backgrounds; Step 3: Under the constraint of minimizing the objective constraint function, update the initial cluster membership degree pixel by pixel to obtain the updated cluster membership degree matrix; Step 4: Update the initial cluster centers based on the cluster membership degree and the probability of belonging to the cluster to obtain the updated cluster centers; Step 5: Set the iteration termination condition based on the previous clustering membership matrix and the current clustering membership matrix. If the iteration termination condition is met or the maximum number of iterations is reached, stop the iteration and output the final clustering membership matrix and the optimal cluster center. Otherwise, repeat steps 3 and 4.

[0062] In one implementation, the target constraint function is constructed based on the attribution probability, including:

[0063] Where J is the objective constraint function, and the smaller the value, the better the clustering effect; Let be the probability of belonging to the j-th pixel. This represents the cluster membership degree of pixel j belonging to the i-th cluster, where m is a preset weighting index and C is a preset number of clusters. Denotes the cluster center of the i-th cluster. Represents pixel grayscale features To the i-th cluster center The squared Euclidean distance represents the gray-level similarity, and N represents the total number of pixels in the initial image.

[0064] In one implementation, the initial cluster membership degree is updated pixel by pixel to obtain the updated cluster membership degree matrix, including:

[0065] in, This represents the clustering membership degree after t iterations, where i is the category number, j is the pixel number, and k is the index. This represents the cluster center of the i-th class obtained after the (t-1)-th iteration. When t=1, This represents the initial cluster centers; Constraints: The sum of the cluster membership degrees of all categories for a single pixel is 1; Updated cluster membership The clustering membership matrix is ​​obtained by filling the matrix according to the position of the i-th row and j-th column.

[0066] In one implementation, the cluster membership degree of each pixel category is updated iteratively based on the previous round of cluster centers. Then, the updated cluster membership degree is combined with the pixel belonging probability weight to solve for new cluster centers. The two work alternately and iteratively to continuously correct the degree of pixel category belonging and the position of each type of cluster center, continuously minimize the weighted distance error of the objective constraint function, correct the deviation caused by the initial random partitioning, and suppress interference such as metal reflection and scattered bright spots in the computer room based on the spatial probability weight in the early stage. This allows the pixel classification and cluster center distribution to continuously conform to the real characteristics of the indicator lights and the background of the cabinet, so that the algorithm gradually converges to the optimal clustering state.

[0067] In one implementation, the initial cluster centers are updated based on cluster membership degree and belonging probability, including:

[0068] in, This represents the i-th cluster center obtained after the t-th iteration.

[0069] In one implementation, the Euclidean distance between the previous clustering membership matrix and the current clustering membership matrix is ​​calculated. If the Euclidean distance is less than the convergence threshold or the number of iterations reaches the maximum number of iterations, the iteration is stopped.

[0070] In one implementation, the pre-defined clustering algorithm integrates the membership, non-membership, and hesitation probabilities as spatial weights into the target constraint function, thereby achieving an organic combination of grayscale uncertainty and spatial distribution. It also introduces scene prior constraints to strengthen the weight of contiguous indicator light areas and weaken the weight of isolated noise points. Furthermore, it adapts to the unique scenarios of computer room halo blur and complex interference, significantly improving segmentation accuracy.

[0071] In one implementation, existing clustering algorithms rely solely on single pixel grayscale features for category classification. This approach fails to adapt to the blurred grayscale transitions at the edges of indicator lights in communication equipment rooms, and it is also difficult to effectively distinguish between real indicator lights and pseudo-targets such as metal reflections from cabinets and isolated, scattered bright spots. Lacking constraints from pixel spatial correlation and fuzzy uncertainty information, it is prone to missegmentation and rough boundaries. Therefore, it is necessary to introduce spatial weights based on previously obtained membership probabilities to construct an improved objective constraint function. By iteratively updating the cluster membership degree and cluster center, the algorithm continuously seeks optimization, integrates pixel grayscale features, visual blur attributes, and prior information from spatial neighborhoods, minimizes weighted distance errors, eliminates reflective noise interference, and achieves smooth and accurate clustering segmentation of indicator light targets and cabinet backgrounds.

[0072] In one embodiment, segmentation processing based on the final cluster membership matrix and the optimal cluster centers is performed to obtain the final image, including: The indicator light target cluster category and the computer room background cluster category are divided according to the optimal cluster center. The membership degree judgment threshold is set by combining the membership degree values ​​of each pixel in the final cluster membership degree matrix. Pixels whose cluster membership degree is greater than the membership degree determination threshold and belong to the target cluster category of the indicator light are marked as foreground pixels of the indicator light, and the remaining pixels are marked as background pixels of the computer room. The marked pixels are subjected to connected component noise reduction to obtain a complete and continuous connected component of the indicator light. The pixels in the connected component of the indicator light are then reconstructed and fused with the background pixels at the pixel level to generate the final image.

[0073] In one implementation, a membership determination threshold is set by combining the membership values ​​of each pixel in the final clustering membership matrix with the corresponding categories, including: Combining the membership value distribution characteristics of the indicator light target cluster and the computer room background cluster corresponding to each pixel in the final membership matrix, we first calculate the mean, median and dispersion of the membership of the corresponding pixels of the two clusters. After removing outliers, we adaptively set a threshold based on the intersection threshold of the membership of the two clusters, combined with the grayscale difference between the bright area of ​​the indicator light and the background and the minimum bright area pixel ratio of the indicator light in the actual inspection scenario. This ensures that the threshold can accurately distinguish the foreground pixels and background pixels of the indicator light, avoid the missed detection of the bright area of ​​the indicator light due to the threshold being too high, and prevent the background noise interference due to the threshold being too low.

[0074] In one implementation, before fusion, adaptive morphological opening and closing operations are applied to the connected domains of the indicator lights to smooth the edges and fill the holes, correct the outline distortion of the indicator lights, and fill in the broken areas inside the bright areas.

[0075] In one implementation, the foreground of the indicator light and the background of the computer room are accurately distinguished by the optimal cluster center and membership threshold, avoiding misjudgment by fixed threshold segmentation; connected component denoising can eliminate isolated noise points, retain the complete and continuous indicator light area, so that the reconstructed final image retains the original brightness and texture details, and removes background interference, significantly improving the robustness and accuracy of indicator light area extraction, providing high-quality input for subsequent state analysis models, thereby enhancing the reliability and real-time performance of inspection and early warning.

[0076] In one embodiment, the state analysis model is based on an improvement of YOLOv8, which includes the following improvements: All C2f modules in the backbone and neck networks of YOLOv8 are replaced with target C2f modules. The target C2f modules are used to enhance the flow and reuse of gradient information in the complex environment of the computer room and improve the ability to extract key features of indicator light status.

[0077] In one implementation, see [link to implementation details]. Figure 2 , Figure 2 This is a schematic diagram of the network structure of the state analysis model provided in the embodiment of the present invention. Based on the characteristics of the module's enhanced gradient information flow and reuse, it effectively resists environmental interference such as complex lighting, equipment obstruction, cluttered background, and dense cabinets in the computer room, accurately captures the key features of the indicator light status in the image, significantly improves the feature extraction accuracy and model recognition robustness of indicator light status detection, and ensures the accuracy and stability of indicator light status detection in complex computer room environments.

[0078] In one implementation, the dataset used to train the state analysis model in the experiment is the detectarlamp dataset; the software environment for training is: operating system: Windows 11, CUDA version: Cuda 12.1, programming language: Python 3.9, deep learning framework: PyTorch 2.0.1; In one implementation, comparative experiments with different models yielded the following results:

[0079] The table clearly shows that, when compared to popular models in existing technologies, the state analysis model is slightly inferior to YOLOv8 and YOLOv11 in terms of latency, but it is significantly better than these two models in terms of mAP75(%) and Params / M. Therefore, the state analysis model is superior to these two models overall.

[0080] In one embodiment, the target C2f module is an improvement upon the C2f module, including: Replace the DarknetBottleneck module in the C2f module with the target Darknet module; The working principle of the target Darknet module is as follows: Obtain the first input feature tensor, split the input feature tensor in half according to the channel dimension to obtain the first feature tensor and the second feature tensor, input the first feature tensor into the CBS module and two concatenated target modules in sequence to obtain the third feature tensor, and input the second feature tensor into the CBS module to obtain the fourth feature tensor; The third and fourth feature tensors are concatenated and then input into the CBS module to obtain the fifth feature tensor. The working principle of the target module is as follows: Obtain the second input feature tensor, and input the second input feature tensor into two parallel Conv modules and one BN module respectively to obtain the first convolutional feature tensor, the second convolutional feature tensor and the third convolutional feature tensor; The first and second convolutional feature tensors are input into two BN modules respectively, and the fourth and fifth convolutional feature tensors are output. The 3rd, 4th, and 5th convolutional feature tensors are concatenated and then sequentially input into the Conv, BN, and CBS modules to output the 6th convolutional feature tensor.

[0081] In one implementation, see [link to implementation details]. Figure 3 , Figure 3 This is an architecture diagram of the target Darknet module provided in this embodiment of the invention. After replacing DarknetBottleneck in the C2f module with this target Darknet module, its internal channel splitting-dual-path parallel processing structure can split the input features by channel. One path passes through CBS and the serial target module to enhance the flow and reuse of gradient information and deeply extract key features of the indicator lights. The other path directly passes through CBS to retain the original feature information. Then, feature complementarity is achieved through splicing and fusion, avoiding the feature loss and gradient vanishing problems of a single processing path. Moreover, while controlling the amount of computation, it enhances the model's ability to capture the status features of indicator lights under complex lighting and background interference in the computer room, significantly improving detection robustness and recognition accuracy, and ensuring the accuracy and stability of indicator light status detection in complex environments. It should be noted that the CBS module consists of Conv, BN (batch normalization layer), and SiLU (activation function).

[0082] In one implementation, see [link to implementation details]. Figure 4 , Figure 4 This is an architecture diagram of the target module provided in an embodiment of the present invention. The target module performs differential processing on the input feature tensor through a three-way parallel multi-branch structure. One path directly passes through the BN module to retain the original feature information, while the two Conv+BN branches extract and enhance the multi-dimensional features related to the indicator light status from different paths. This enriches the feature expression and strengthens gradient flow and alleviates the gradient vanishing problem through multi-path parallelism. The stable feature distribution of each branch and the fused BN module improves training stability and generalization ability. The fused Conv and CBS modules further integrate and refine multi-path features, ultimately achieving accurate capture of key features of indicator lights under complex lighting and background interference in the computer room.

[0083] In one embodiment, the indicator light status category output by the status analysis model is obtained; the status category includes a normal state and at least one abnormal state, and the abnormal state includes an indicator light not lit, an indicator light always lit error, an abnormal indicator light flashing frequency, and an indicator light color error; When the status category is normal, continue to execute the next round of image acquisition and inspection; When the state category is an abnormal state, extract the anomaly type and confidence score from the analysis results; Preset warning levels are matched based on the anomaly type and confidence level. The warning levels include Level 1, Level 2, and Level 3. Execute the corresponding warning operation according to the warning level; the warning operation includes at least one of the following: generate an anomaly log and store it in the database, highlight the location of the anomaly indicator light in the monitoring interface and display the status details, send an alarm message to the operation and maintenance personnel terminal, or trigger the on-site audible and visual alarm. If the same indicator light repeatedly triggers the same level of warning within the preset continuous inspection cycle, the warning level will be automatically upgraded and higher-level maintenance personnel will be notified.

[0084] In one implementation, a level one warning corresponds to a complete failure of the indicator light, a level two warning corresponds to an error in color or flashing frequency, and a level three warning corresponds to a decrease in brightness or a slight abnormality.

[0085] One implementation method precisely categorizes indicator lights into normal and various abnormal states, matches corresponding warning levels based on the anomaly type and confidence level, and executes diverse warning operations to achieve accurate anomaly identification and rapid response. Simultaneously, a warning escalation mechanism is added for repeated triggering of the same anomaly within a continuous inspection cycle, effectively preventing fault omissions, improving the targeting and efficiency of maintenance, reducing manual inspection errors, lowering the risk of equipment failures in communication equipment rooms caused by indicator light anomalies, ensuring the stable and reliable operation of the equipment room, and providing intelligent and efficient support for equipment room maintenance.

[0086] The foregoing has described one embodiment of the present invention in detail, but this content is merely a preferred embodiment and should not be considered as limiting the scope of the present invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the scope of the claims of this invention.

Claims

1. A machine vision-based intelligent inspection and early warning method for indicator light status in communication equipment rooms, characterized in that, The method includes: The original images of the indicator lights in the communication equipment room are acquired, and the original images are preprocessed to obtain an initial image; Extract the core bright area image of the computer room indicator light from the initial image, and perform feature extraction based on the initial image to obtain fuzzy features; the fuzzy features include optimized membership degree, non-membership degree, and hesitation degree; The probability of each pixel belonging to an indicator light is calculated based on the core bright area image and blur features. The final clustering membership matrix and optimal cluster center of the indicator light are obtained by a preset clustering algorithm based on the membership probability. The final image is obtained by segmentation based on the final clustering membership matrix and the optimal cluster center. The final image is input into the state analysis model to obtain the analysis results of the indicator light status, and an early warning process is performed based on the analysis results. The state analysis model is based on an improvement of YOLOv8, and the improvement includes: replacing all C2f modules in the backbone network and neck network of YOLOv8 with target C2f modules. The target C2f modules are used to improve the extraction capability of key features of the indicator light status.

2. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, Preprocessing the original image to obtain an initial image includes: The original image is converted into a grayscale image, and the brightness value and spatial coordinates of the target pixel are extracted; the target pixel is any pixel in the grayscale image. The membership and non-membership of the target pixel are initialized based on the brightness value of the target pixel; The brightness value, spatial coordinates, membership degree and non-membership degree of the target pixel are bound and fused to generate an information-containing target pixel. The information-containing target pixel is then filled into the corresponding position in the original image to obtain the initial image.

3. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, Extracting the core bright area image of the server room indicator lights from the initial image, including: Perform brightness saliency detection processing on the initial image to output a brightness saliency image of the indicator light; The first pixel is obtained by filtering each row of pixels in the bright image by using a brightness threshold, and the area where the first pixel is located is used as the first candidate indicator area. The first brightness data of the pixels within the first candidate indicator light area is statistically analyzed, and the indicator light row length data and center horizontal coordinate data of the first candidate indicator light area are calculated based on the first brightness data; the first brightness data includes the sum of the brightness of the pixels within the first candidate indicator light area, the maximum brightness value, and the total number of pixels; The core pixel of the indicator light in each row is determined based on the row length data and center horizontal coordinate data of the candidate indicator light area in each row. The core pixels of each row of indicator lights are then stitched together to obtain the indicator light row image. The second pixel is obtained by filtering each column of pixels in the bright image by using a brightness threshold, and the area where the second pixel is located is used as the second candidate indicator area. The second brightness data of the pixels within the second candidate indicator area is statistically analyzed, and the indicator length data and center vertical coordinate data of the second candidate indicator area are calculated based on the second brightness data; the second brightness data includes the sum of the brightness of the pixels within the second candidate indicator area, the maximum brightness value, and the total number of pixels; The core pixel of the indicator light in each column is determined based on the column length data and center vertical coordinate data of the candidate indicator light area of ​​each column of pixels. The core pixels of each column of indicator lights are then stitched together to obtain the indicator light column image. The row image and column image of the indicator lights are fused together to obtain the core bright area image of the computer room indicator lights.

4. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 3, characterized in that, Based on the initial image, feature extraction is performed to obtain blurred features, including: The brightness threshold is calculated, and the brightness boundary is used to segment each pixel in the initial image to obtain the optimized membership degree of each pixel belonging to the indicator light. The brightness boundary includes a low brightness boundary and a high brightness boundary. The non-membership degree and hesitation degree of each pixel are calculated using a preset algorithm based on the optimized membership degree of each pixel.

5. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, The probability of each pixel belonging to an indicator light is calculated based on the core bright area image and blur features, including: Based on the initial image, all pixels in the core region of the core bright area image are selected and a valid pixel set is formed. Taking any pixel in the initial image as the center, select pixels within an 8-neighborhood, count the number of pixels within the neighborhood that belong to the effective pixel set, and calculate the neighborhood weight factor based on the adjustment coefficient. The probability of each pixel belonging to an indicator light is calculated based on the fuzzy features and the neighborhood weight factor.

6. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, Based on the membership probability, a pre-defined clustering algorithm is used to obtain the final clustering membership matrix and optimal cluster centers for the indicator lights, including: Step 1: Randomly select a preset number of pixel grayscale features in the initial image as the initial cluster centers, and initialize the initial cluster membership. Step 2: Construct the target constraint function based on the membership probability; the target constraint function is used to find the optimal cluster center and cluster assignment membership degree to achieve accurate fuzzy clustering of indicator light targets and complex backgrounds; Step 3: Under the constraint of minimizing the objective constraint function, update the initial cluster membership degree pixel by pixel to obtain the updated cluster membership degree matrix; Step 4: Update the initial cluster centers based on the cluster membership matrix and the membership probability to obtain the updated cluster centers; Step 5: Set the iteration termination condition based on the previous clustering membership matrix and the current clustering membership matrix. If the iteration termination condition is met or the maximum number of iterations is reached, stop the iteration and output the final clustering membership matrix and the optimal cluster center. Otherwise, execute steps 3 and 4 to enter the next round of iteration.

7. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, The final image is obtained by segmentation based on the final clustering membership matrix and the optimal cluster centers, including: Based on the optimal cluster center, the indicator light target cluster category and the computer room background cluster category are divided. Combining the membership degree values ​​of each pixel in the final cluster membership degree matrix, the membership degree judgment threshold is set. Pixels whose cluster membership degree is greater than the membership degree determination threshold and belong to the target cluster category of the indicator light are marked as foreground pixels of the indicator light, and the remaining pixels are marked as background pixels of the computer room. The marked pixels are subjected to connected component noise reduction to obtain a complete and continuous connected component of the indicator light. The pixels in the connected component of the indicator light are then reconstructed and fused with the background pixels at the pixel level to generate the final image.

8. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, The state analysis model is based on an improvement of YOLOv8, and the improvements include: All C2f modules in the backbone and neck networks of YOLOv8 are replaced with target C2f modules; the target C2f modules are used to enhance the flow and reuse of gradient information in the complex environment of the computer room and improve the ability to extract key features of indicator light status.

9. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 8, characterized in that, The target C2f module is an improvement upon the C2f module, including: Replace the DarknetBottleneck module in the C2f module with the target Darknet module; The working principle of the target Darknet module is as follows: Obtain the first input feature tensor, split the input feature tensor in half according to the channel dimension to obtain the first feature tensor and the second feature tensor, input the first feature tensor into the CBS module and two concatenated target modules in sequence to obtain the third feature tensor, and input the second feature tensor into the CBS module to obtain the fourth feature tensor; The third and fourth feature tensors are concatenated and then input into the CBS module to obtain the fifth feature tensor. The working principle of the target module is as follows: Obtain the second input feature tensor, and input the second input feature tensor into two parallel Conv modules and one BN module respectively to obtain the first convolutional feature tensor, the second convolutional feature tensor and the third convolutional feature tensor; The first and second convolutional feature tensors are input into two BN modules respectively, and the fourth and fifth convolutional feature tensors are output. The 3rd, 4th, and 5th convolutional feature tensors are concatenated and then sequentially input into the Conv, BN, and CBS modules to output the 6th convolutional feature tensor.

10. The intelligent inspection and early warning method for indicator light status in a communication equipment room based on machine vision according to claim 1, characterized in that, Based on the analysis results, early warning processing is carried out, including: Obtain the indicator light status category output by the status analysis model; the status category includes normal status and at least one abnormal status, the abnormal status includes indicator light not lit, indicator light always lit error, indicator light flashing frequency abnormal, indicator light color error; When the status category is normal, continue to execute the next round of image acquisition and inspection; When the state category is an abnormal state, the abnormality type and confidence score are extracted from the analysis results. A preset warning level is matched based on the anomaly type and confidence level value. The warning levels include Level 1 warning, Level 2 warning, and Level 3 warning. Execute the corresponding warning operation according to the warning level; the warning operation includes at least one of the following: generate an abnormal log and store it in the database, highlight the abnormal indicator light position on the monitoring interface and display the status details, send an alarm message to the operation and maintenance personnel terminal, and trigger the on-site audible and visual alarm; If the same indicator light repeatedly triggers the same level of warning within the preset continuous inspection cycle, the warning level will be automatically upgraded and higher-level maintenance personnel will be notified.

Images