Subway protection area illegal construction identification and detection method based on image identification

Abstract

The invention discloses a subway protection area illegal construction recognition and detection method based on image recognition, and particularly relates to the technical field of image recognition. According to the method, target equipment is detected and classified by collecting and analyzing construction scene image data in a subway protection area, meanwhile, feature extraction and perception range limitation degree evaluation are performed according to spectral characteristics of camouflage materials, and whether the equipment has camouflage behaviors or not is accurately judged in combination with equipment construction plan data fluctuation conditions. Once the camouflage behavior is recognized, the system prompts the constructor to stop operation through the linkage sound-light alarm device, uploads the characteristics of the camouflage material to the database, dynamically optimizes the spectrum sensing model of the camera, improves the detection capability of the camera on the novel camouflage behavior, effectively solves the problem of monitoring failure caused by the camouflage material and a sensing blind area, and improves the monitoring efficiency. The accuracy of illegal construction detection in the subway protection area and the safety and reliability of the system are greatly improved, and therefore stable operation of subway facilities is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and in particular to a method for identifying and detecting illegal construction in a subway protection zone based on image recognition. Background Art

[0002] Illegal construction identification and detection in subway protection zones based on image recognition refers to the use of image recognition technology to monitor and judge possible illegal construction activities within the subway protection zone, and to achieve all-weather, high-precision automated supervision through electronic sentinels. Subway protection zones usually refer to areas within a certain range around subway lines, where construction activities may pose a threat to the safety and stability of subway facilities. As intelligent monitoring nodes, electronic sentinels can automatically detect unauthorized or non-compliant construction activities through real-time collection and analysis of on-site images or videos, and proactively issue warnings or initiate corresponding protective measures. For example, intelligent electronic sentinel cameras are installed at a construction site in a subway protection zone. These cameras can not only capture high-definition images around the clock, but also have intelligent analysis and edge computing capabilities. They transmit data to the image recognition system in real time and identify specific construction behaviors such as excavation equipment and hoisting operations through algorithms. If the electronic sentinel finds that an unapproved excavator operation exists in a certain area, it will immediately mark it as an abnormality and automatically send an alarm signal to the management department, while recording the time and specific location of the behavior for quick positioning and processing. This electronic sentinel technology combined with image recognition algorithms not only significantly reduces the cost of manual inspections, but also significantly improves the efficiency and accuracy of construction supervision within protected areas, effectively reducing the operational risks of subway facilities.

[0003] The prior art has the following deficiencies: Construction parties may effectively avoid detection by smart electronic sentinel cameras by covering the surface of equipment with special materials so that it only reflects light of specific wavelengths (such as infrared or ultraviolet). These materials can reduce the characteristic reflection of the equipment in the visible light band, or change its appearance texture, making it impossible for image recognition algorithms to effectively extract key features. For example, an excavator may look like an ordinary object, or become blurred in the surveillance video. In addition, if the spectral perception range of the smart electronic sentinel camera is limited and it cannot detect changes in spectral characteristics caused by camouflage materials, illegal construction equipment may be "invisible" in the monitoring system, making it impossible for the system to locate and identify the camouflage behavior of key equipment. This will directly lead to the failure to timely detect illegal construction activities in the subway protection zone, seriously weaken the reliability and safety of the monitoring system, and may eventually cause major safety hazards such as damage to subway facilities and interruption of operations. Summary of the invention

[0004] The purpose of the present invention is to provide a method for identifying and detecting illegal construction in a subway protection zone based on image recognition, so as to overcome the shortcomings of the background technology.

[0005] In order to achieve the above object, the present invention provides the following technical solution: a method for identifying and detecting illegal construction in a subway protection zone based on image recognition, comprising the following steps: S1: Intelligent electronic sentinel cameras are deployed in the subway protection area to continuously collect image data of the construction scene and record the timestamp and geographic location information of the image during collection; S2: Preprocess the collected image data and transmit the preprocessed image to the image recognition system, analyze the preprocessed image data based on the deep learning algorithm, detect and classify the target construction equipment in the construction scene; S3: Combine the multi-spectral response capability of the intelligent electronic sentinel camera with the spectral reflectance characteristics of the camouflage material to analyze the spectral characteristics of the target construction equipment and evaluate the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera; S4: Determine whether the target construction equipment has camouflage behavior based on the limited degree of the spectral perception range of the intelligent electronic sentinel camera and the fluctuation of the construction plan data of the target construction equipment; S5: For identified camouflage behaviors, the management platform links the sound and light alarm devices on site to remind construction workers to stop the operation immediately, and uploads the spectral characteristics of the identified camouflage materials to the spectral reflectance characteristics database to optimize the camera spectral perception model.

[0006] Preferably, in S3, after analyzing the multi-spectral response capability of the intelligent electronic sentinel camera, a camera spectral response anomaly index is generated, and the method for obtaining the camera spectral response anomaly index is: Collect the camera's spectral response data matrix X, where each row represents the spectral response curve of a different camera at each wavelength; standardize each column of the matrix X: ; is the standardized data, is the mean of the jth column, is the standard deviation of the jth column, Represents the element in the matrix X, located at the i-th row and j-th column, and calculates the covariance matrix C of the standardized data matrix Z: ; Where T is the matrix transpose, is an n×n matrix, which represents the covariance between bands; the covariance matrix C is decomposed by eigenvalue to obtain the eigenvalue and the corresponding eigenvector ; The decomposition expression is: ; Eigenvalue Represents the variance contribution of the kth principal component, the eigenvector represents the direction of the kth principal component; Project the standardized data Z onto the first p principal components: ; Y is the representation of the data in the principal component space, with a size of m×p, is a matrix containing the first p eigenvectors, with a size of n×p; the camera spectral response anomaly index of each sample in the principal component space is calculated, and the expression is: ; Where DFC is the camera spectral response anomaly index, is the principal component vector of the i-th sample, with a size of 1×p, is the mean vector of all samples in the principal component space, with a size of 1×p, and S is the covariance matrix in the principal component space, with a size of p×p.

[0007] Preferably, in S3, the spectral reflectance characteristics of the camouflage material are analyzed to generate a spectral reflectance deviation index, and the spectral reflectance deviation index is obtained by: Assume that the spectral reflectance curve of the camouflage material is ,in represents the reflectivity at the wth wavelength; let the reference standard spectral reflectivity curve be ,in represents the reflectivity at the mth wavelength; Define the local distance of spectral reflectance at a certain wavelength: ; The local distance between the spectral reflectance curve Q of the camouflage material at the ath wavelength and the reference spectral reflectance curve R at the bth wavelength is constructed to construct a cumulative distance matrix D of size w×m, where each element D(a,b) represents the minimum cumulative distance from the starting point to the current point, and the calculation formula is: The boundary conditions are ; v is an index used to represent the accumulation process in the cumulative distance calculation, starting from the lower right corner of the matrix D(w,m), backtracking along the path with the smallest cumulative distance until reaching the starting point D(1,1): The path is recorded as , where L is the path length; the spectral reflectance deviation value is calculated based on the cumulative distance D(w,m) of the cumulative path P, and the expression is: ; Where EDH is the spectral reflectance deviation index.

[0008] Preferably, in S3, the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is evaluated, specifically: the camera spectral response anomaly index and the spectral reflectance deviation index are converted into a comprehensive feature vector, and the comprehensive feature vector is used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the score value label of the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera as the prediction target, and takes minimizing the sum of prediction errors of the score value labels of the degree of limitation of the spectral perception range of all intelligent electronic sentinel cameras as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The score value of the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0009] Preferably, the obtained score value of the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the score value of the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is compared with the first standard threshold and the second standard threshold respectively; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than the second standard threshold, it means that the spectral perception range limitation of the intelligent electronic sentinel camera is high, and a first-level warning signal is generated at this time; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the spectral perception range limitation of the intelligent electronic sentinel camera is moderate, and a second-level warning signal is generated at this time; If the score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is less than the first standard threshold, it means that the spectral perception range limitation degree of the intelligent electronic sentinel camera is low, and a third-level warning signal is generated at this time.

[0010] Preferably, in S4, the time deviation between the actual construction behavior of the target construction equipment and the predetermined construction plan is analyzed to generate a construction time deviation value, and the method for obtaining the construction time deviation value is: Get the actual construction time series, represented as a vector ,in, is the starting or ending time point of the construction equipment in the qth time period; Get the scheduled construction plan time series, represented as a vector ,in is the starting or ending time point of the qth planned time period, and the Euclidean distance between the actual construction time vector and the scheduled construction plan time vector is calculated: ; Where D is the construction time deviation value in the qth planned time period; if the construction behavior is divided into multiple task segments, the deviation needs to be calculated separately for each task segment and accumulated: ; In the formula, is the construction time deviation value, g is the total number of task segments, is the number of time points in the hth task segment.

[0011] Preferably, the spectral perception range limitation score Lw of the intelligent electronic sentinel camera and the construction time deviation value As the input item of fuzzy logic, whether the target construction equipment has disguised behavior is taken as the output item. The judgment results of the disguised behavior of the target construction equipment include: no disguised behavior, possible disguised behavior and clear disguised behavior; Define fuzzy sets of input and output terms; Use triangular or trapezoidal membership functions to represent the membership of input and output fuzzy sets; According to the relationship between input items and output items, a fuzzy rule base is established; The output of each rule is weighted according to the activation strength; Convert fuzzy results into clear output values; According to the defuzzification results , determine whether the target construction equipment has camouflage behavior: : No disguised behavior; : There may be disguised behavior; : There is clear evidence of disguise.

[0012] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention realizes comprehensive monitoring and precise analysis of construction equipment in subway protection areas by combining intelligent electronic sentinel cameras, multi-spectral response capabilities, deep learning algorithms and fuzzy logic. The camera collects image data and detects and classifies the target construction equipment. Combined with the evaluation of the degree of limitation of the spectral perception range and the analysis of fluctuations in construction plan data, it can quickly determine whether the construction equipment has camouflaged behavior. Once camouflaged behavior is identified, the system generates a graded early warning signal and promptly reminds construction personnel to stop operating through an audible and visual alarm device. At the same time, the spectral characteristics of the camouflaged material are uploaded to dynamically optimize the detection model, thereby improving the system's ability to detect new camouflaged behaviors.

[0013] 2. The present invention effectively solves the problems of blind monitoring areas and limited spectral perception caused by camouflage materials in the prior art, significantly improves the accuracy and efficiency of detecting illegal construction behaviors in subway protection areas, and reduces the safety risks of subway facilities. At the same time, the analysis method combining machine learning and fuzzy logic is adopted to make the system highly adaptable and flexible, able to cope with complex and changing construction environments, and provide intelligent and scalable technical support for the safety management of subway protection areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0015] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0017] For examples, see Figure 1 As shown, the method for identifying and detecting illegal construction in a subway protection zone based on image recognition described in this embodiment includes the following steps: S1: Intelligent electronic sentinel cameras are deployed in the subway protection area to continuously collect image data of the construction scene and record the timestamp and geographic location information of the image during collection; S2: Preprocess the collected image data and transmit the preprocessed image to the image recognition system, analyze the preprocessed image data based on the deep learning algorithm, detect and classify the target construction equipment in the construction scene; S3: Combine the multi-spectral response capability of the intelligent electronic sentinel camera with the spectral reflectance characteristics of the camouflage material to analyze the spectral characteristics of the target construction equipment and evaluate the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera; S4: Determine whether the target construction equipment has camouflage behavior based on the limited degree of the spectral perception range of the intelligent electronic sentinel camera and the fluctuation of the construction plan data of the target construction equipment; S5: For identified camouflage behaviors, the management platform links the sound and light alarm devices on site to remind construction workers to stop the operation immediately, and uploads the spectral characteristics of the identified camouflage materials to the spectral reflectance characteristics database to optimize the camera spectral perception model.

[0018] In S1, based on the scope of the subway protection zone, the distribution characteristics of construction activities and risk assessment, intelligent electronic sentinel cameras are rationally deployed to ensure coverage of high-risk areas (such as above subway tunnels, around pipelines, etc.). The density of camera deployment is determined so that the monitoring areas of adjacent cameras overlap seamlessly to avoid monitoring blind spots.

[0019] Smart camera features include: High-definition imaging capability: Equipped with high-definition lenses, it supports high-resolution (such as 4K or above) image and video acquisition to ensure clear details. Multi-spectral perception capability: Integrates visible light, infrared and ultraviolet spectrum perception modules to identify devices or behaviors that may use camouflage materials. Dynamic zoom function: With automatic zoom and viewing angle adjustment functions, it can track dynamic construction targets in real time. Edge computing capability: Integrates edge computing modules to process part of the image data locally to reduce transmission delays.

[0020] The camera continuously collects real-time image data of the construction scene, supporting all-weather monitoring, including nighttime infrared imaging and stable acquisition in adverse weather conditions. The built-in clock module automatically adds an accurate timestamp to each frame of the image. The built-in GPS module or through regional calibration automatically records the geographic location information of the image acquisition point to ensure the accuracy of event tracking and positioning.

[0021] The collected data is transmitted to the backend management platform in real time via 5G network or dedicated wireless network. Encrypted transmission protocols (such as TLS / SSL) are used to protect data security and prevent data from being tampered with or intercepted during transmission. In the event of a network interruption, the camera can cache data locally and automatically upload it after the network is restored. The camera has a built-in motion detection module. When it detects that construction equipment enters the protected area, it automatically activates image acquisition and marks related events. Combined with the early warning mechanism, higher frequency image data can be collected first when the target equipment enters a sensitive area.

[0022] In this application, through the above layout and functional configuration, the intelligent electronic sentinel camera can efficiently collect construction scene data in the subway protection zone, and achieve accurate positioning and real-time monitoring of construction behavior through timestamp and geographic location information marking. This design provides basic data support for subsequent image analysis and abnormal behavior detection.

[0023] S2: Preprocess the collected image data and transmit the preprocessed image to the image recognition system. Analyze the preprocessed image data based on the deep learning algorithm to detect and classify the target construction equipment in the construction scene.

[0024] Preprocess the collected image data to improve image quality, optimize the data format for input models, and adapt to complex construction scene requirements, including: Remove random noise from the image through median filtering, mean filtering, or adaptive filtering algorithms, especially in images collected in low-light or dusty environments. Use histogram equalization or brightness enhancement algorithms to optimize images with insufficient lighting conditions to ensure that equipment edges and details are clearly visible. Enhance the saliency of construction equipment in complex backgrounds through contrast stretching or adaptive contrast enhancement techniques. Uniformly adjust the image to the input size of the deep learning model (such as 416×416 or 224×224 pixels) to ensure the efficiency and consistency of model calculations. Convert the image from RGB format to grayscale images or other color spaces (such as HSV or LAB) to highlight equipment features according to model requirements. Generate diverse data samples by slightly rotating, mirroring, and randomly cropping the image to enhance the robustness of the model. Introduce actual noise in the simulated scene (such as blur, interference lines, etc.) into the image to improve the model's performance in real environments.

[0025] Transmit the pre-processed image to the image recognition system to ensure efficient and secure data transmission. Losslessly compress the image data to reduce transmission bandwidth while maintaining image quality. Use high-speed communication networks (such as 5G or optical fiber) to achieve low-latency real-time transmission. Protect the integrity and confidentiality of image data during transmission through encryption protocols (such as AES or RSA).

[0026] Analyze the preprocessed image data based on the deep learning algorithm to complete the detection and classification of the target construction equipment, including: using target detection models (such as YOLO, Faster R-CNN, SSD, etc.) to detect the equipment in the construction scene. Input: preprocessed image. Output: detection box (Bounding Box) and the location coordinates of the target equipment. Detect multiple devices in the same scene at the same time to ensure comprehensive identification of equipment in complex construction sites. Through multi-frame continuous detection combined with optical flow algorithm, the motion trajectory of the equipment is tracked to provide support for subsequent behavior analysis.

[0027] Use classification models (such as ResNet, Inception, EfficientNet, etc.) to classify the detected target equipment. Classification categories: Excavation equipment (such as excavators, drilling rigs). Lifting equipment (such as cranes, tower cranes). Transport equipment (such as trucks, tracked vehicles). Use deep feature extraction modules (such as convolutional layer outputs) to extract key features of the target equipment and compare them with the authorized equipment features in the construction plan database.

[0028] Output the category, location, and size of the detected device. Add a confidence score to each recognition result to indicate the model's confidence in the classification. Overlay the recognition results (such as annotation boxes and category labels) on the original image to help managers intuitively understand the recognition results.

[0029] The detection and classification results are stored in the system database as input data for subsequent behavior analysis and camouflage detection. If an unauthorized device is detected, the system alarm mechanism is triggered and the pre-processed image and analysis results are sent to the management platform.

[0030] For example, by combining the YOLOv5 target detection model, the optical flow algorithm, and the ResNet50 classification model, real-time detection, classification, and motion trajectory tracking of excavators, trucks, and cranes in construction scenes can be achieved. Specifically: Input data: Video frames captured by the camera are preprocessed (noise reduction, enhancement, and normalization) and then input into the target detection model.

[0031] Single-frame example image: The image contains an excavator, a truck, and a crane, with a complex construction scene in the background.

[0032] Use YOLOv5 (You Only Look Once) as the target detection model.

[0033] The model performs feature extraction and convolution processing on the image to generate detection results.

[0034] Output the detection box (Bounding Box) of each target device and its position coordinates in the image: Excavator: Bounding Box = [50, 100, 200, 300]. Truck: Bounding Box = [220, 120, 350, 300]. Crane: Bounding Box = [400, 50, 550, 250]. The model successfully detects the three devices in the image and generates the detection box and position coordinates of each device.

[0035] Combine target detection and optical flow algorithm. Perform correlation analysis on multiple consecutive frames of detection results and track device position changes through optical flow algorithm.

[0036] Output the motion trajectory of each device: Excavator: moves from [50,100] to [55,105]. Truck: moves from [220,120] to [230,130]. ​​Crane: fixed position, not moving. The motion trajectory shows that the excavator and truck are moving, while the crane remains stationary, which is consistent with the actual situation of the construction scene.

[0037] ResNet50 (Residual Network) is used as the classification model. For each detected target device, an image sub-region is extracted from the detection box area. A convolutional neural network (CNN) is used to extract deep features and classify them: Excavator: Classification result = "Excavating equipment" (98% confidence). Truck: Classification result = "Transportation equipment" (95% confidence). Crane: Classification result = "Lifting equipment" (96% confidence). The model accurately classifies the categories of excavator, truck, and crane, all with high confidence.

[0038] Get the feature records of authorized devices from the construction plan database: Authorized device category: Excavator (equipment number: EX001, feature: yellow paint). Truck (equipment number: TR002, feature: blue paint). Crane (equipment number: CR003, feature: red paint). Compare the detected device features: The detected excavator matches the authorized device EX001 (the paint is consistent and the features match). The detected truck matches the authorized device TR002 (the paint is consistent and the features match). The detected crane does not match any authorized device (the paint is green and the features are abnormal). The excavator and truck are authorized devices, and the crane is not in the authorized list and may be an unauthorized device.

[0039] Equipment detection and classification results: Excavators, trucks, and cranes were successfully detected and classified. Unauthorized equipment (green crane) was identified. The system alerted the on-site personnel to stop the crane operation through the sound and light alarm device. The image, location, and unmatched feature information of the crane were uploaded to the management platform. The system stored the feature data of the crane (green paint, crane equipment category) in the database to optimize the model for future camouflage behavior detection.

[0040] S3: Combining the multi-spectral response capability of the intelligent electronic sentinel camera with the spectral reflectance characteristics of the camouflage material, the spectral characteristics of the target construction equipment are analyzed to evaluate the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera.

[0041] Generate monochromatic light in different wavelength ranges (for example, 300-1100nm, including ultraviolet, visible light, and near-infrared) through a standard light source (such as a spectrometer). Test the imaging effect and signal strength of the intelligent electronic sentinel camera at each wavelength, and record the spectral response curve. Obtain the sensitivity curve of the camera in the ultraviolet, visible light, and near-infrared ranges, indicating the wavelength range and blind spots of the perception capability. Collect images under different spectral conditions in actual construction scenarios to test the spectral perception effect, such as infrared monitoring capabilities at night and imaging clarity in visible light environments during the day. Evaluation indicators include: Multispectral coverage: whether the wavelength range that the camera can effectively perceive meets the detection requirements. Pixel accuracy: the resolution and quality of imaging in the edge wavelength range.

[0042] After analyzing the multi-spectral response capability of the intelligent electronic sentinel camera, the camera spectral response anomaly index is generated. The method for obtaining the camera spectral response anomaly index is as follows: Collect the camera's spectral response data matrix X, where each row represents the spectral response curve of different cameras at each wavelength, with a size of m×n, where m is the number of samples (measurement data from different cameras or at different times). n is the number of spectral bands (for example, the number of points sampled from 300nm to 1100nm).

[0043] Standardize each column of the matrix X (i.e. the data of each band) so that its mean is 0 and its standard deviation is 1: ; Z is the standardized data, is the mean of the jth column, is the standard deviation of the jth column, Represents the element in the matrix X, located at the i-th row and j-th column, and calculates the covariance matrix C of the standardized data matrix Z: ; Where T is the matrix transpose, is an n×n matrix, which represents the covariance between bands; the covariance matrix C is decomposed by eigenvalue to obtain the eigenvalue and the corresponding eigenvector ; The decomposition expression is: ; Eigenvalue Represents the variance contribution of the kth principal component, the eigenvector represents the direction of the kth principal component; Project the standardized data Z onto the first p principal components (usually select the number of principal components that makes the cumulative variance contribution reach 90%-95%): ; Y is the representation of the data in the principal component space, with a size of m×p, is a matrix containing the first p eigenvectors, of size n×p; Calculate the camera spectral response anomaly index of each sample in the principal component space. The expression is: ; Where DFC is the camera spectral response anomaly index, is the principal component vector of the i-th sample, with a size of 1×p, is the mean vector of all samples in the principal component space, with a size of 1×p, and S is the covariance matrix in the principal component space, with a size of p×p.

[0044] The larger the camera spectral response anomaly index is, the more obvious the spectral perception range of the smart electronic sentinel camera is, and the response of certain bands deviates greatly from the normal spectral characteristics. This usually means that the camera's perception ability in these bands is severely limited. For example, it cannot effectively detect the absorption or reflection bands commonly used by camouflage materials (such as near-infrared or ultraviolet), making the camouflage equipment "invisible". This anomaly will directly reduce the camera's recognition accuracy of construction equipment and increase the difficulty of detecting illegal construction.

[0045] On the contrary, the smaller the camera's spectral response anomaly index is, the more comprehensive its spectral perception range is and the response is consistent with the expected response, without obvious blind spots. This shows that the camera can effectively perceive the target characteristics in most bands, and can accurately identify the target through multi-spectral data even in the face of interference from camouflage materials. Therefore, cameras with smaller anomaly indexes have higher adaptability and monitoring reliability in complex scenes.

[0046] Collect samples of camouflage materials that may be used by the construction party, including infrared absorption coatings, ultraviolet reflection films, and other special coatings. Use a spectrometer to measure the reflectivity curve of each camouflage material in the range of 300-1100nm. Record the reflection enhancement or absorption characteristics of the camouflage material in a specific wavelength range. Coat the surface of the model equipment (such as a small excavator) with camouflage materials and conduct multispectral imaging tests in a controlled environment. Compare the images of the camouflaged equipment captured by the intelligent electronic sentinel camera with the images of the normal equipment to analyze the camouflage effect. Store the spectral reflectance characteristics of the camouflage material obtained from the test in the spectral database to form a camouflage feature model for subsequent comparison.

[0047] After analyzing the spectral reflectance characteristics of the camouflage material, a spectral reflectance deviation index is generated. The method for obtaining the spectral reflectance deviation index is as follows: Assume that the spectral reflectance curve of the camouflage material is ,in represents the reflectivity at the wth wavelength; let the reference standard spectral reflectivity curve be ,in represents the reflectivity at the mth wavelength; Define the local distance of spectral reflectance at a certain wavelength: ; The local distance between the spectral reflectance curve Q of the camouflage material at the ath wavelength and the reference spectral reflectance curve R at the bth wavelength is constructed to construct a cumulative distance matrix D of size w×m, where each element D(a,b) represents the minimum cumulative distance from the starting point to the current point, and the calculation formula is: The boundary conditions are ; v is an index used to represent the accumulation process in the cumulative distance calculation, starting from the lower right corner of the matrix D(w,m), backtracking along the path with the smallest cumulative distance until reaching the starting point D(1,1): The path is recorded as , where L is the path length; the spectral reflectance deviation value is calculated based on the cumulative distance D(w,m) of the cumulative path P, and the expression is: ; Where EDH is the spectral reflectance deviation index.

[0048] The larger the spectral reflectance deviation index, the more significant the deviation between the spectral reflectance curve of the camouflage material and the reference standard. This shows that the perception ability of the intelligent electronic sentinel camera in certain key spectral bands is severely limited and fails to effectively capture the true characteristics of the camouflage material. For example, the camouflage material may show abnormal reflectance changes in the camera blind area (such as the near-infrared or ultraviolet bands), but the camera fails to respond, making it difficult for the monitoring system to identify the camouflage behavior. In this case, the camera's spectral perception range is highly limited, threatening the reliability of monitoring.

[0049] The smaller the spectral reflectance deviation index, the closer the spectral reflectance curve of the camouflage material is to the reference standard, and the camera can accurately capture the spectral characteristics of the camouflage material. This shows that the intelligent electronic sentinel camera has good perception capabilities in multiple spectral bands and can detect subtle features and changes in reflective characteristics of camouflage materials, thereby effectively avoiding interference of camouflage behavior on the monitoring system. In this case, the camera's spectral perception range is less restricted, and the monitoring system is more reliable and adaptable.

[0050] The camera's spectral response anomaly index and spectral reflectance deviation index are converted into comprehensive feature vectors, which are used as inputs to the machine learning model. The machine learning model predicts the spectral perception range limitation degree score label of the intelligent electronic sentinel camera with each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of prediction errors of the spectral perception range limitation degree score labels of all intelligent electronic sentinel cameras as the training target. The machine learning model is trained until the sum of prediction errors converges, and the model training is stopped. The spectral perception range limitation degree score of the intelligent electronic sentinel camera is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0051] The method for obtaining the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is as follows: from the comprehensive feature vector training data of the trained machine learning model, the corresponding function expression is obtained: ; In the formula, is the output function of the model, DFC is the camera spectral response anomaly index, EDH is the spectral reflectance deviation index, A score indicating the degree of limitation of the spectral perception range of the smart electronic sentinel camera.

[0052] The obtained score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is compared with the first standard threshold and the second standard threshold respectively; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than the second standard threshold, it means that the spectral perception range of the intelligent electronic sentinel camera is highly limited and may not be able to detect the camouflage behavior of the target construction equipment. In this case, a first-level warning signal is generated; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the spectral perception range limitation of the intelligent electronic sentinel camera is moderate, the camera perception range is partially limited, and some camouflage behaviors may be missed. At this time, a second-level warning signal is generated; If the score of the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is less than the first standard threshold, it means that the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is low, the perception range covers comprehensively, and the interference to camouflage behavior is small. At this time, a third-level warning signal is generated.

[0053] S4: Determine whether the target construction equipment is camouflaged based on the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera and the fluctuation of the construction plan data of the target construction equipment.

[0054] The fluctuation of the construction plan data of the target construction equipment refers to the deviation between the actual construction behavior of the equipment and the scheduled construction plan, including changes in multiple dimensions such as location, time, and equipment type. For example, whether a certain equipment is working in an unauthorized area, whether it exceeds the scheduled construction time window, or whether the type of equipment used does not match the type filed in the plan. These fluctuations may reflect abnormalities or potential violations of construction activities, and are an important basis for determining whether the equipment is disguised or unauthorized.

[0055] The time deviation between the actual construction behavior of the target construction equipment and the scheduled construction plan is analyzed to generate a construction time deviation value. The method for obtaining the construction time deviation value is as follows: Get the actual construction time series, represented as a vector ,in, It is the starting or ending time point of the construction equipment in the qth time period.

[0056] Get the scheduled construction plan time series, represented as a vector ,in is the starting or ending time point of the qth planned time period. Calculate the Euclidean distance between the actual construction time vector and the scheduled construction plan time vector: ; Where D is the construction time deviation value in the qth planned time period.

[0057] If the construction activity is divided into multiple task segments (e.g. different construction areas or equipment), the deviation needs to be calculated separately for each task segment and accumulated: ; In the formula, is the construction time deviation value, g is the total number of task segments, is the number of time points in the hth task segment.

[0058] The spectral perception range limitation score Lw and the construction time deviation value of the intelligent electronic sentinel camera are used as the input items of fuzzy logic, and whether the target construction equipment has camouflage behavior is used as the output item. The judgment results of the camouflage behavior of the target construction equipment include: no camouflage behavior, possible camouflage behavior and clear camouflage behavior.

[0059] Defining fuzzy sets includes: Input fuzzy sets: Spectral perception range limitation score Lw: Low, Medium, High. Construction time deviation value : Small, Medium, Large.

[0060] Output fuzzy set: Disguise behavior judgment results: No disguise, Possible disguise, Definite disguise.

[0061] Use triangular or trapezoidal membership functions to represent the membership of input and output fuzzy sets.

[0062] Example: Spectral perception range limitation rating Lw: ; ; ; Construction time deviation The membership functions are defined similarly, for example: Small: Trapezoidal membership function, range [0,10], center [0,5].

[0063] Medium: triangular membership function, range [5,20], peak value 12.5.

[0064] Large: Trapezoidal membership function, range [15,∞], start 20.

[0065] According to the relationship between input items and output items, a fuzzy rule base is established. For example: When the spectral sensing range Lw is low (Low) and When it is Small, the target construction equipment is judged as having no disguise.

[0066] When Lw is Medium and When it is Small, the target construction equipment may be disguised (Possible disguise).

[0067] When Lw is High and When it is Small, the target construction equipment may be disguised (Possible disguise).

[0068] When Lw is High and When the value is Medium, the target construction equipment is clearly disguised (Definite disguise).

[0069] When Lw is High and When it is Large, the target construction equipment is clearly disguised (Definite disguise).

[0070] According to the input Lw and , and calculate the activation strength of each rule (taking the minimum value of the input item membership).

[0071] The output of each rule is weighted according to the activation strength.

[0072] Using the centroid method, convert the fuzzy result into a clear output value: ; In the formula, is the activation strength of rule i, is the output value corresponding to rule i (for example, 0 means no disguise, 0.5 means possible disguise, and 1 means clear disguise).

[0073] According to the defuzzification results , determine whether the target construction equipment has camouflage behavior: : No disguised behavior; : There may be disguised behavior; : There is clear evidence of disguise.

[0074] S5: For identified camouflage behaviors, the management platform links the sound and light alarm devices on site to remind construction workers to stop the operation immediately, and uploads the spectral characteristics of the identified camouflage materials to the spectral reflectance characteristics database to optimize the camera spectral perception model.

[0075] When the intelligent electronic sentinel camera detects camouflage behavior and determines the severity of the camouflage behavior through fuzzy logic or scoring mechanism, the management platform immediately sends out an alarm signal.

[0076] The sound and light alarm device includes a tweeter and a high-brightness flashing light, which respectively issue the following warnings: Sound alarm: Continuously or intermittently play voice prompts, such as "If you find disguised behavior, please stop construction operations immediately." Light alarm: Use red or yellow flashing lights to attract the attention of construction workers to avoid safety risks caused by continued construction due to disguised behavior. Strong warning signals force on-site construction workers to stop illegal construction and ensure the safety of the subway protection area.

[0077] The management platform extracts the reflective characteristics of camouflage materials from the images and spectral data collected by the camera, which usually include the following: Key band range: For example, the reflection or absorption characteristics of camouflage materials in the infrared, ultraviolet or visible light bands. Spectral curve characteristics: Standardize the spectral reflectance curve of the material to form characteristic data.

[0078] The extracted spectral characteristic data is uploaded to the spectral reflectance characteristic database to store the detailed characteristics of the camouflage material for subsequent detection and analysis. The data storage fields include: the band characteristics of the camouflage material. The record of the time and place of the material. The response data during camera detection.

[0079] The database is updated dynamically, and as more camouflage material characteristic data is entered, the database is gradually expanded. Each update marks the version of the database so that the latest camouflage material characteristic data can be selected when the model is called.

[0080] Combined with the characteristics of camouflage materials uploaded to the database, the camera's spectral perception model is updated to optimize the detection ability of camouflage materials: Model training data enhancement: Use the characteristics of new camouflage materials as training data, retrain the model, and improve the model's recognition ability for new camouflage materials. Update the spectral matching algorithm in the model so that it can quickly detect behaviors similar to the characteristics of newly added camouflage materials.

[0081] If the camera has edge computing capabilities, the optimized spectral perception model is sent to the camera via the wireless network to enhance the camera's local detection capabilities. After optimization, the camera can: Improve sensitivity to specific spectral ranges (such as infrared and ultraviolet). More accurately identify spectral anomalies of camouflage materials. Regularly iterate the spectral perception model to ensure that the camera can adapt to the rapid changes in camouflage materials, such as the emergence of new coatings or materials with special reflective properties.

[0082] In this embodiment, in the subway protection zone, intelligent electronic sentinel cameras are deployed to continuously collect image data of the construction scene, and the timestamp and geographic location information are recorded. The collected image data is transmitted to the image recognition system after preprocessing, and the target equipment in the construction scene is detected and classified using a deep learning algorithm. Combining the multi-spectral response capability of the camera and the spectral reflectance characteristics of the camouflage material, the spectral characteristics of the target equipment are analyzed to evaluate the degree of limitation of the camera's spectral perception range. According to the degree of limitation of the camera's perception range and the fluctuation of the construction plan data of the target equipment, it is judged whether there is camouflage behavior. If camouflage behavior is detected, the management platform links the on-site sound and light alarm device to remind the construction personnel to stop the operation, and uploads the spectral characteristics of the camouflage material to the database to optimize the camera's spectral perception model and improve the detection capability and reliability of the system.

[0083] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A method for identifying and detecting illegal construction in a subway protection zone based on image recognition, characterized in that: The following steps are involved: S1: Intelligent electronic sentinel cameras are deployed in the subway protection area to continuously collect image data of the construction scene and record the timestamp and geographic location information of the image during collection; S2: Preprocess the collected image data and transmit the preprocessed image to the image recognition system, analyze the preprocessed image data based on the deep learning algorithm, detect and classify the target construction equipment in the construction scene; S3: Combine the multi-spectral response capability of the intelligent electronic sentinel camera with the spectral reflectance characteristics of the camouflage material to analyze the spectral characteristics of the target construction equipment and evaluate the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera; S4: Determine whether the target construction equipment has camouflage behavior based on the limited degree of the spectral perception range of the intelligent electronic sentinel camera and the fluctuation of the construction plan data of the target construction equipment; S5: For identified camouflage behaviors, the management platform links the sound and light alarm devices on site to remind construction workers to stop the operation immediately, and uploads the spectral characteristics of the identified camouflage materials to the spectral reflectance characteristics database to optimize the camera spectral perception model.

2. According to the method for identifying and detecting illegal construction in a subway protection zone based on image recognition in claim 1, it is characterized by: In S3, the multi-spectral response capability of the intelligent electronic sentinel camera is analyzed to generate the camera spectral response anomaly index. The method for obtaining the camera spectral response anomaly index is as follows: Collect the camera's spectral response data matrix X, where each row represents the spectral response curve of a different camera at each wavelength; standardize each column of the matrix X: ; is the standardized data, is the mean of the jth column, is the standard deviation of the jth column, Represents the element in matrix X, located at the i-th row and j-th column, and calculates the covariance matrix C of the standardized data matrix Z: ; Where T is the matrix transpose, is an n×n matrix, which represents the covariance between bands; the covariance matrix C is decomposed by eigenvalue to obtain the eigenvalue and the corresponding eigenvector ; The decomposed expression is: ; Eigenvalue Represents the variance contribution of the kth principal component, the eigenvector represents the direction of the kth principal component; Project the standardized data Z onto the first p principal components: ; Y is the representation of the data in the principal component space, with a size of m×p, is a matrix containing the first p eigenvectors, of size n×p; Calculate the camera spectral response anomaly index of each sample in the principal component space. The expression is: ; Where DFC is the camera spectral response anomaly index, is the principal component vector of the i-th sample, with a size of 1×p, is the mean vector of all samples in the principal component space, with a size of 1×p, and S is the covariance matrix in the principal component space, with a size of p×p.

3. The method for identifying and detecting illegal construction in a subway protection zone based on image recognition according to claim 2 is characterized in that: In S3, the spectral reflectance characteristics of the camouflage material are analyzed to generate a spectral reflectance deviation index. The spectral reflectance deviation index is obtained as follows: Assume that the spectral reflectance curve of the camouflage material is ,in represents the reflectivity at the wth wavelength; Assume that the reference standard spectral reflectance curve is ,in represents the reflectivity at the mth wavelength; Define the local distance of spectral reflectance at a certain wavelength: ; The local distance between the spectral reflectance curve Q of the camouflage material at the ath wavelength and the reference spectral reflectance curve R at the bth wavelength is constructed to construct a cumulative distance matrix D of size w×m, where each element D(a,b) represents the minimum cumulative distance from the starting point to the current point, and the calculation formula is: The boundary conditions are ; v is an index used to represent the accumulation process in the cumulative distance calculation, starting from the lower right corner of the matrix D(w,m), backtracking along the path with the smallest cumulative distance until reaching the starting point D(1,1): the path is recorded as , where L is the path length; the spectral reflectance deviation value is calculated based on the cumulative distance D(w,m) of the cumulative path P, and the expression is: ; Where EDH is the spectral reflectance deviation index.

4. The method for identifying and detecting illegal construction in a subway protection zone based on image recognition according to claim 3 is characterized in that: In S3, the degree of limitation of the spectral perception range of the intelligent electronic sentinel camera is evaluated, specifically: the camera spectral response anomaly index and spectral reflectance deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as inputs of the machine learning model. The machine learning model predicts the spectral perception range limitation degree score value label of the intelligent electronic sentinel camera for each group of comprehensive feature vectors as the prediction target, and minimizes the sum of prediction errors of the spectral perception range limitation degree score value labels of all intelligent electronic sentinel cameras as the training target. The machine learning model is trained until the sum of prediction errors converges and the model training is stopped. The spectral perception range limitation degree score value of the intelligent electronic sentinel camera is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

5. The method for identifying and detecting illegal construction in a subway protection zone based on image recognition according to claim 4 is characterized in that: The obtained score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is compared with the first standard threshold and the second standard threshold respectively; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than the second standard threshold, it means that the spectral perception range limitation of the intelligent electronic sentinel camera is high, and a first-level warning signal is generated at this time; If the score value of the spectral perception range limitation of the intelligent electronic sentinel camera is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the spectral perception range limitation of the intelligent electronic sentinel camera is moderate, and a second-level warning signal is generated at this time; If the score value of the spectral perception range limitation degree of the intelligent electronic sentinel camera is less than the first standard threshold, it means that the spectral perception range limitation degree of the intelligent electronic sentinel camera is low, and a third-level warning signal is generated at this time.

6. The method for identifying and detecting illegal construction in a subway protection zone based on image recognition according to claim 1, characterized in that: In S4, the time deviation between the actual construction behavior of the target construction equipment and the predetermined construction plan is analyzed to generate a construction time deviation value. The method for obtaining the construction time deviation value is: Get the actual construction time series, represented as a vector ,in, is the starting or ending time point of the construction equipment in the qth time period; Get the scheduled construction plan time series, represented as a vector ,in is the starting or ending time point of the qth planned time period, and the Euclidean distance between the actual construction time vector and the scheduled construction plan time vector is calculated: ; Where D is the construction time deviation value in the qth planned time period; if the construction behavior is divided into multiple task segments, the deviation needs to be calculated separately for each task segment and accumulated: ; In the formula, is the construction time deviation value, g is the total number of task segments, is the number of time points in the hth task segment.

7. The method for identifying and detecting illegal construction in a subway protection zone based on image recognition according to claim 6 is characterized in that: The spectral perception range limitation score Lw and construction time deviation value of the intelligent electronic sentinel camera are As the input item of fuzzy logic, whether the target construction equipment has disguised behavior is taken as the output item. The judgment results of the disguised behavior of the target construction equipment include: no disguised behavior, possible disguised behavior and clear disguised behavior; Define fuzzy sets of input and output terms; Use triangular or trapezoidal membership functions to represent the membership of input and output fuzzy sets; According to the relationship between input items and output items, a fuzzy rule base is established; The output of each rule is weighted according to the activation strength; Convert fuzzy results into clear output values; According to the defuzzification results , determine whether the target construction equipment has camouflage behavior: : No disguised behavior; : There may be disguised behavior; : There is clear evidence of disguise.

Images