Track foreign matter intelligent detection system and method based on image recognition

Abstract

The invention discloses an intelligent track foreign matter detection system and method based on image recognition, and particularly relates to the technical field of image recognition. A real-time image is acquired, an area-of-interest is extracted in combination with an image segmentation algorithm, and potential foreign matters are identified; the space-time consistency of the detection result is verified by analyzing the space-time data of the track foreign matter so as to judge the real threat; extracting multi-dimensional features such as shape, texture, color and size from a real threat target, and outputting classification confidence to evaluate model classification stability; the detection accuracy is evaluated by combining the space-time consistency and the classification stability, the detection results are divided into different categories, the abnormal degree of subsequent detection is predicted for the incomplete accuracy detection result, and the image acquisition resolution or frame rate is dynamically adjusted to improve the detection definition. The problems of missing detection and frequent false alarm of small and medium-sized foreign matters in the prior art are solved, the accuracy and reliability of track foreign matter detection are greatly improved, and the running safety of a high-speed train is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and particularly to an intelligent detection system and method for track foreign objects based on image recognition. Background Art

[0002] Intelligent detection of track foreign objects based on image recognition refers to using image recognition technology to analyze images on the track through intelligent algorithms, so as to identify and determine whether there are foreign objects on the track that may affect traffic safety. This technology usually combines high-resolution cameras, sensors, and artificial intelligence models (such as deep learning) to achieve real-time monitoring and anomaly detection, ensuring the smoothness and safety of the track. For example, in high-speed railway or subway systems, cameras installed along the track can continuously collect track images. When a foreign object (such as a dropped tool, gravel, branch, or other obstacles) enters the track, the system will quickly identify these anomalies through a trained deep learning model and trigger an alarm to notify relevant staff to clean up the site or take emergency measures. For instance, during the operation of a railway in a certain place, the system detected a dropped steel plate on the track and reported it in a timely manner through the image recognition model, avoiding an accident that might have caused a train derailment.

[0003] The prior art has the following deficiencies: When the intelligent detection system detects foreign objects with small volume or colors similar to the track background (such as bolts in rail rust), it may fail due to high target concealment or limited camera angles. Although these small foreign objects are not large in size, they may pose significant safety hazards during the operation of high-speed trains, such as the risk of derailment. For example, a black bolt dropped on a dark rail may be misidentified as a normal track by the system due to its similar texture to the background and thus missed. In addition, if the image recognition algorithm is inaccurate in detecting foreign objects and wrongly identifies harmless objects (such as leaves, plastic bags) as dangerous foreign objects, resulting in frequent alarms of the system, it will not only increase the maintenance cost, but also reduce the trust of the staff in the alarms, thus ignoring real threats. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent detection system and method for track foreign objects based on image recognition to solve the deficiencies in the background art.

[0005] To achieve the above purpose, the present invention provides the following technical solution: An intelligent detection method for track foreign objects based on image recognition, including the following steps: S1: Real-time obtain image data of the track area through high-definition cameras set along the track, and preprocess the obtained image data, including denoising, contrast enhancement, and color correction; S2: Use an image segmentation algorithm to segment the track area, extract the regions of interest within the track area, and perform object detection on the extracted regions of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; S3: Obtain the spatio-temporal data of the track foreign objects in different frame images, analyze the position changes and movement directions of the track foreign objects in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection results, and determine whether the detection target is a real threat; S4: After determining that the detection target is a real threat, extract the multi-dimensional features of the detected foreign objects, including shape, texture, color, and size, output the classification confidence through the deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm; S5: Evaluate the accuracy of the image recognition algorithm for detecting track foreign objects based on the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm. According to the evaluation results, divide the track foreign object detection results into accurate detection results, partially accurate detection results, and inaccurate detection results; S6: For the partially accurate detection results, predict the degree of abnormality in the accuracy of the image recognition algorithm for detecting track foreign objects during the subsequent detection time period. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

[0006] Preferably, in S2, use an image segmentation algorithm to segment the track area, extract the railway track edges and straight-line structures through edge detection and Hough transformation, and distinguish between the left and right sides of the track; use a semantic segmentation model to classify each pixel in the image, separate the track area from the background area, and label the pixel tags of the track area for the model to learn the track features; based on the segmentation results, retain the image part of the track area; Define the region of interest ROI within the track area, delimit the ROI boundary through the geometric characteristics of the track width and spacing; use a convolutional neural network model to perform object detection on the extracted ROI region, select the input track area image and ROI label, combine the labeled foreign object positions and categories for model training, and optimize the model using classification loss and regression loss; output the bounding box coordinates, classification labels, and confidence levels of the foreign objects.

[0007] Preferably, in S3, after analyzing the position changes and movement directions of the track foreign objects in consecutive frames, generate a foreign object spatio-temporal consistency index. The method for obtaining the foreign object spatio-temporal consistency index is as follows: Set the target trajectory sequence as , representing the position sequence of the detected track foreign objects in consecutive frames, and set the reference trajectory sequence , representing the standard motion pattern of the track foreign object; the Euclidean distance is used to calculate the distance between two trajectory points , the expression is: ; where is the position coordinate of the target trajectory point, is the position coordinate of the reference trajectory point, and the cumulative distance matrix D is constructed: ; where represents the cumulative minimum distance from the starting point of the trajectory to point, is the Euclidean distance between the current points, represents the optimal path selection from the previous position to the current point, and the boundary conditions are: ; ; ; In the cumulative distance matrix, find the optimal path P from the starting point (1,1) to the end point (m,n), and the path consists of a series of points that satisfy: ; represents selecting the path P to minimize the sum of the cumulative distances of all points on the path, and calculating the spatio-temporal consistency index of the foreign object, the expression is: ; In the formula, L is the length of the optimal path, is the Euclidean distance on the path, and HAK is the spatio-temporal consistency index of the foreign object.

[0008] Preferably, in S4, after analyzing the proportion of low-confidence targets, a low-confidence target proportion fluctuation index is generated, and the acquisition method of the low-confidence target proportion fluctuation index is: Collect the time series of the proportion of low-confidence targets ; represents the proportion of low-confidence targets in the T-th frame image, and T is the total number of frames in the time series. The time series R is segmented and classified into w intervals: ; represents the boundary of the w-th interval; count the number of data points in each interval , the expression is: ; Calculate the probability of the i-th interval: ; is the probability that the low-confidence proportion falls into the i-th interval, and calculate the low-confidence target proportion fluctuation index, the expression is: ; In the formula, GXH is the low-confidence target proportion fluctuation index.

[0009] Preferably, in S5, according to the spatio-temporal consistency of the track foreign object detection result and the classification stability of the deep learning model algorithm, evaluate the accuracy of the image recognition algorithm for track foreign object detection; Convert the foreign object spatio-temporal consistency index and the low-confidence target proportion fluctuation index into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes the prediction of the accuracy value label of the image recognition algorithm for detecting track foreign objects for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of all image recognition algorithms for detecting track foreign objects as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the accuracy value of the image recognition algorithm for detecting track foreign objects according to the model output result, where the machine learning model is a polynomial regression model.

[0010] Preferably, in S5, according to the evaluation result, divide the track foreign object detection result into accurate detection result, incomplete accurate detection result and inaccurate detection result, specifically: Compare the obtained accuracy value of the image recognition algorithm for detecting track foreign objects with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the accuracy value of the image recognition algorithm for detecting track foreign objects with the first standard threshold and the second standard threshold respectively; If the accuracy value of the image recognition algorithm for detecting track foreign objects is greater than the second standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is high. At this time, generate a high-accuracy detection signal and divide the track foreign object detection result into an accurate detection result; If the accuracy value of the image recognition algorithm for detecting track foreign objects is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is medium. At this time, generate a medium-accuracy detection signal and divide the track foreign object detection result into an incomplete accurate detection result; If the accuracy value of the image recognition algorithm for detecting track foreign objects is less than the first standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is low. At this time, generate a low-accuracy detection signal and divide the track foreign object detection result into an inaccurate detection result.

[0011] Preferably, in S6, for the incomplete accurate detection result, predict the abnormal degree of the accuracy of the image recognition algorithm for detecting track foreign objects in the subsequent detection time period, specifically: For the incomplete accuracy detection results, that is, the accuracy value of the image recognition algorithm for detecting track foreign objects generated within a fixed detection time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, collect the accuracy values greater than or equal to the first standard threshold and less than or equal to the second standard threshold within the subsequent detection time period, establish the corresponding data set, calculate the mean and standard deviation of the data set, and after analyzing it, predict the abnormal degree of the accuracy of the image recognition algorithm for detecting track foreign objects within the subsequent detection time period according to the analysis results.

[0012] Preferably, if the mean value of the accuracy values in the data set is greater than or equal to the reference threshold of the mean value of the accuracy values, and the standard deviation of the accuracy values is less than the reference threshold of the standard deviation of the accuracy values, the detection result is stable and the accuracy is high. At this time, no warning signal is generated, no adjustment is required, and the existing detection settings can be maintained; If the mean value of the accuracy values is greater than or equal to the reference threshold of the mean value of the accuracy values, and the standard deviation of the accuracy values is greater than or equal to the reference threshold of the standard deviation of the accuracy values, the detection accuracy fluctuates greatly. At this time, a third-level warning signal is generated, and the algorithm parameters are adjusted or data is increased; If the mean value of the accuracy values is less than the reference threshold of the mean value of the accuracy values, and the standard deviation of the accuracy values is greater than or equal to the reference threshold of the standard deviation of the accuracy values, the detection accuracy is low and the fluctuation is significant. At this time, a first-level warning signal is generated, and improvement measures should be taken, including optimizing the model or improving the image acquisition quality; If the mean value of the accuracy values is less than the reference threshold of the mean value of the accuracy values, and the standard deviation of the accuracy values is less than the reference threshold of the standard deviation of the accuracy values, the detection accuracy is low. At this time, a second-level warning signal is generated, and the model should be optimized or the algorithm should be switched.

[0013] The present invention also provides an intelligent detection system for track foreign objects based on image recognition, including an image acquisition and preprocessing module, an image segmentation and target detection module, a spatio-temporal consistency analysis module, a classification stability evaluation module, a detection result evaluation and classification module, and a dynamic optimization and anomaly prediction module; Image acquisition and preprocessing module: Real-time obtain the image data of the track area through high-definition cameras set along the track, and preprocess the obtained image data, including denoising, contrast enhancement, and color correction; Image segmentation and target detection module: Use image segmentation algorithms to segment the track area, extract the regions of interest within the track area, and perform target detection on the extracted regions of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; Spatio-temporal Consistency Analysis Module: Obtain the spatio-temporal data of the track foreign object in different frame images, analyze the position changes and movement directions of the track foreign object in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection result, and determine whether the detection target is a real threat; Classification Stability Evaluation Module: After determining that the detection target is a real threat, extract the multi-dimensional features of the detected foreign object, including shape, texture, color, and size, output the classification confidence through a deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm; Detection Result Evaluation and Classification Module: According to the spatio-temporal consistency of the track foreign object detection result and the classification stability of the deep learning model algorithm, evaluate the accuracy of the image recognition algorithm for detecting track foreign objects. According to the evaluation results, divide the track foreign object detection results into accurate detection results, incomplete accurate detection results, and inaccurate detection results; Dynamic Optimization and Anomaly Prediction Module: For incomplete accurate detection results, predict the degree of abnormality in the accuracy of the image recognition algorithm for detecting track foreign objects during subsequent detection time periods. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention: 1. The present invention combines an image segmentation algorithm and a deep learning model to achieve efficient detection and classification of foreign objects in the track area. Aiming at the problems of high concealment of small foreign objects and complex track backgrounds, by introducing a spatio-temporal consistency index of foreign objects and a fluctuation index of the proportion of low-confidence targets, comprehensively analyze the spatio-temporal characteristics and classification stability of the detection target, and improve the detection accuracy of small targets and targets similar to the background. The polynomial regression model further converts the multi-dimensional feature vector into a detection accuracy prediction value, combined with dynamically adjusting the image acquisition resolution or frame rate, effectively coping with the fluctuation problem of detection performance, and ensuring the real-time performance and adaptability of the system.

[0015] 2. The present invention can greatly reduce the problems of missed detection of small foreign objects and false alarms of harmless objects, and improve the accuracy and reliability of track foreign object detection. Based on the accuracy evaluation, classify the detection results, and through the early warning mechanism, timely identify abnormal detection situations and provide targeted optimization strategies to ensure the long-term stable operation of the system. By dynamically optimizing image acquisition and the deep learning model, the adaptability of the system to complex track environments is improved, the potential safety hazards and operation and maintenance costs are significantly reduced, providing a solid guarantee for the safe operation of rail transit. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other accompanying drawings can also be obtained based on these drawings.

[0017] Figure 1 It is the method flowchart of the present invention.

[0018] Figure 2 It is the system module diagram of the present invention. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0020] Embodiment 1. Please refer to Figure 1 As shown, a method for intelligent detection of track foreign objects based on image recognition in this embodiment includes the following steps: S1: Real-time obtain image data of the track area through high-definition cameras set along the track, and preprocess the obtained image data, including denoising, contrast enhancement, and color correction; S2: Use an image segmentation algorithm to segment the track area, extract the regions of interest within the track area, and perform object detection on the extracted regions of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; S3: Obtain the spatio-temporal data of the track foreign objects in different frame images, analyze the position changes and movement directions of the track foreign objects in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection results, and determine whether the detection target is a real threat; S4: After determining that the detection target is a real threat, extract multi-dimensional features of the detected foreign object, including shape, texture, color, and size, output the classification confidence through the deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm; S5: Evaluate the accuracy of the image recognition algorithm for detecting track foreign objects according to the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm. According to the evaluation results, divide the track foreign object detection results into accurate detection results, incomplete accurate detection results, and inaccurate detection results; S6: For the incomplete accuracy detection results, predict the degree of abnormality in the accuracy of the image recognition algorithm for detecting track foreign objects during the subsequent detection time period. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

[0021] In S1, use a high-resolution camera (such as 4K or higher resolution) to ensure that track details can be captured, such as the outlines and textures of small foreign objects. Select a camera with a wide dynamic range (WDR) to adapt to complex lighting conditions (such as shadows and direct sunlight). Install the cameras at intervals along the track to ensure coverage of the entire track area. Adjust the camera angles to ensure a panoramic view of the track area is captured and avoid blind spots. The acquisition frequency should match the train running speed. For example, for high-speed railways, 30 frames per second or higher can be selected to ensure that track details are not missed.

[0022] Preprocess the images to remove noise generated by sensors or environmental factors and improve the image quality. Common noise types: Gaussian noise: caused by random variations in the camera sensor. Lighting noise: from reflections or uneven light sources in the surrounding environment. Gaussian filtering: reduce high-frequency noise through the Gaussian smoothing algorithm while preserving image edge information. Median filtering: suitable for removing salt-and-pepper noise (such as bright or dark spots) and protecting the sharpness of the track edges. Non-local Means (NL-Means) filtering: perform weighted averaging on similar pixel blocks to effectively denoise while maintaining details.

[0023] Improve the clarity of the images in the track area to make target foreign objects (such as bolts and gravel) more obvious in the images. There may be low contrast in the track area, such as insufficient lighting or a too complex background. Enhance the contrast of the brightness in the image by redistributing pixel intensity values. This is applicable when the overall brightness distribution in the track area is uniform but the details are not obvious. Contrast Limited Adaptive Histogram Equalization (CLAHE): process the image in blocks to enhance local contrast and avoid noise amplification caused by over-enhancement. This is applicable when the track background is complex and the lighting is uneven. Gamma correction: enhance dark details or reduce overexposure in highlight areas by adjusting the gamma value (non-linear transformation). It is commonly used to optimize track details in shadow areas.

[0024] Correct the color deviation in the image to make the image closer to the real scene and ensure that the color characteristics of foreign objects (such as rust and plastic bags) can be displayed correctly. The camera may cause image color distortion under different light conditions (such as daylight, shadow, and night lights). White balance correction: Adjust the white point in the image to make the image color distribution more natural. For example, by adjusting the ratio of RGB channels, the color deviation caused by different light sources can be corrected. Color mapping: Use a reference color card to map the color values in the image to a standard color space. Applicable to track monitoring environments with color reference benchmarks. Color smoothing: Apply a filtering algorithm to eliminate local color abnormal areas in the image to make the color distribution of the track and foreign objects more uniform.

[0025] In this application, the high-definition camera collects and performs denoising, contrast enhancement and color correction on the track image in real time, which can greatly improve the image quality and the accuracy of identifying foreign objects, and provide clear and reliable image data for subsequent target detection and classification. These preprocessing steps are the key basic links of the intelligent detection system for foreign objects on the track.

[0026] S2: Segment the track area using an image segmentation algorithm, extract the region of interest within the track area, perform target detection on the extracted region of interest based on a deep learning model, and identify potential track foreign objects, wherein the deep learning model includes a convolutional neural network.

[0027] Use image segmentation algorithms to segment the track area, separate the track area (such as rails and sleepers) from the background area (such as the ground and grass), and clearly mark the track area as the focus of analysis. The specific steps include: Extract the track edge using Canny edge detection or Sobel operator. Detect the straight structure and edge contour of the rail in the preprocessed image. Detect the straight structure of the track using Hough transform. Fit the detected edge points into straight lines to distinguish the left and right sides of the rail. Segment the track area based on region growing or threshold segmentation. Use a semantic segmentation model (such as U-Net or DeepLab) to classify each pixel in the image and distinguish the track area from the background area. Label the pixel labels of the track area during training so that the model can learn the track features. Based on the segmentation results, only the image part of the track area is retained and irrelevant background is removed.

[0028] Extract the region of interest (ROI) within the track area. The region of interest (ROI) refers to the key area in the image that may contain foreign objects on the target track, and the system takes it as the focus of subsequent detection. Examples: the track surface, track joints, and the area between the rail and the sleeper. Combine the segmentation results of the track and use the geometric characteristics of the track width and spacing to define the ROI boundary. For example, set the area within a certain range (such as 10 pixels) on both sides of the rail as the ROI. Dynamically adjust the ROI range according to the track environment. For example, expand the ROI in the track curve area to cover more possible foreign object distribution areas. Analyze the historical detection data of track foreign objects, focus on the areas where foreign objects often appear, and optimize the ROI range.

[0029] Use a deep learning model to perform object detection on the ROI. The selection of convolutional neural network (CNN) models includes: Object detection models: YOLO (You Only Look Once): It has strong real-time detection performance and is suitable for the high real-time requirements of track foreign object detection. Faster R-CNN: A high-precision object detection algorithm, suitable for scenarios that require fine detection. SSD (Single Shot Detector): Balances speed and accuracy and is suitable for multi-object scenarios.

[0030] Collect a large amount of track foreign object data, including various environments (sunny, rainy) and foreign object types (such as bolts, gravel, tools).

[0031] Annotate the positions (bounding boxes) and categories (such as tools, plastic bags) of the foreign objects.

[0032] Input: Track area image and ROI labels.

[0033] Loss function: Use classification loss (cross-entropy) and regression loss (L1 or L2 loss) to optimize the model.

[0034] Data augmentation: Simulate various actual environments through rotation, flipping, and lighting changes to enhance the robustness of the model.

[0035] Use the extracted ROI area as the model input for object detection.

[0036] Output detection results: Position: The bounding box coordinates of the foreign object. Category: The classification label of the foreign object (such as tool, gravel). Confidence: The confidence probability of each detection result, indicating the credibility of the model for the detection result.

[0037] Improve the detection ability for small targets (such as bolts, gravel) through multi-scale feature extraction (such as FPN). Combine the detection confidence and a preset threshold to screen out high-confidence targets as potential foreign objects. For example, if the detection confidence is greater than the preset threshold, then the target is regarded as a potential foreign object; If the same foreign object is detected in multiple-frame images, use the spatio-temporal fusion method to improve the accuracy of the detection results. The output content includes: the position, category, and confidence level of the foreign object, which are visually displayed by overlaying them on the track image. Send an alarm message to the monitoring center and mark the specific dangerous points in the track area.

[0038] S3: Obtain the spatio-temporal data of the track foreign object in different-frame images, analyze the position changes and movement directions of the track foreign object in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection results, and determine whether the detection target is a real threat.

[0039] Obtain the spatio-temporal data of the track foreign object, including: Time data: referring to the timestamps when the foreign object appears in different-frame images. Spatial data: referring to the coordinate positions of the foreign object in the image, including two-dimensional image coordinates (such as x, y) and possible depth information (z, if lidar or stereo cameras are used).

[0040] In consecutive-frame images, perform cross-frame association of the target based on the appearance features (such as shape, texture, color) or position proximity of the target. Use Kalman filtering or the Optical Flow method to track the target movement trajectory. Only retain the detection targets within the track area and exclude the interference from non-track areas (such as background dynamic targets). Record the timestamp, spatial coordinates, and detection confidence level of the target in each frame to form the spatio-temporal data set of the foreign object.

[0041] Analyze the position changes and movement directions of the track foreign object in consecutive frames, and use the coordinate positions of the foreign object in consecutive-frame images to calculate the movement trajectory. If the position change in consecutive frames is close to 0, it is judged as a stationary target (such as a bolt, gravel). If the position change is significantly different and the direction changes continuously over time, it may be a dynamic target (such as a plastic bag). Use trajectory curve fitting (such as the least squares method) to evaluate whether the movement trajectory is smooth and reasonable. Unreasonable situations: sudden changes in position, large jumps in direction. Classify the trajectory characteristics: Stationary trajectory: stable targets on the track, such as dropped tools. Regular trajectory: leaves or plastic bags moving with the wind, with a random but continuous movement trajectory. Abnormal trajectory: sudden changes in the target appearance position or non-compliance with physical laws (possibly a false detection).

[0042] After analyzing the position changes and movement directions of the track foreign object in consecutive frames, generate a spatio-temporal consistency index for the foreign object. Based on the spatio-temporal consistency index of the foreign object, determine whether the detection target is a real threat. The method for obtaining the spatio-temporal consistency index of the foreign object is as follows: Set the target trajectory sequence as , representing the position sequence of the detected track foreign object in consecutive frames (such as two-dimensional coordinates (x, y)), and set the reference trajectory sequence , representing the possible standard movement patterns of the track foreign object (such as stationary, regular movement). Use the Euclidean distance to calculate the distance between two trajectory points , the expression is: ; where is the position coordinate of the target trajectory point, is the position coordinate of the reference trajectory point, and a cumulative distance matrix D is constructed: ; where represents the cumulative minimum distance from the starting point of the trajectory to the point, is the Euclidean distance between the current points, represents the optimal path selection from the previous position to the current point, and the boundary conditions are: ; ; ; In the cumulative distance matrix, find the optimal path P from the starting point (1, 1) to the end point (m, n). The path consists of a series of points that satisfy: ; represents selecting the path P to minimize the sum of the cumulative distances of all points on the path. Calculate the spatio-temporal consistency index of the foreign object, and the expression is: ; In the formula, L is the length of the optimal path, is the Euclidean distance on the path, and HAK is the spatio-temporal consistency index of the foreign object.

[0043] When the spatio-temporal consistency index of the foreign object is small, it indicates that the detected trajectory has a high degree of matching with the reference trajectory, the trajectory is smooth, and the motion law conforms to the actual scenario. For example, a stationary foreign object (such as a bolt or gravel) has almost unchanged positions in multiple consecutive frames on the track, or a slowly moving foreign object (such as a sliding tool) has a linear trajectory and a consistent direction. In this case, the detection target has high spatio-temporal consistency and is determined to be a real threat because such targets usually pose actual potential hazards to the track safety.

[0044] When the spatio-temporal consistency index of the foreign object is large, it indicates that the degree of matching between the trajectory and the reference trajectory is low, the trajectory is not smooth or there are abnormal changes. This usually indicates that the detected target is a dynamic interference object (such as a plastic bag fluttering in the wind or a fast-moving shadow), and its trajectory may show irregular jumps, sudden changes in direction, or drastic changes in speed. In this case, the detection target has low spatio-temporal consistency and is not determined to be a real threat, and the alarm can be excluded to avoid false alarms.

[0045] S4: After determining that the detection target is a real threat, extract the multi-dimensional features of the detected foreign object, including shape, texture, color, and size, output the classification confidence through the deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm.

[0046] After determining that the detection target is a real threat, multi-dimensional features of the detected foreign object are extracted, including shape features: describing the geometric structure of the foreign object, such as edges, contours, and corner points. Shape feature vectors (such as roundness, aspect ratio) of the target are extracted through Hough transform, edge detection (such as Canny or Sobel). Texture features include: describing the details and structural patterns on the surface of the foreign object, such as roughness, smoothness. Local Binary Pattern (LBP), Gray-Level Co-Occurrence Matrix (GLCM), etc. are used to calculate the texture features of the foreign object. Color features include: describing the color distribution of the foreign object and its contrast with the background. Color histograms are extracted to analyze the distribution of each channel in RGB, HSV, or Lab color spaces. Size features include: describing the physical size and occupied area of the foreign object. Size features (such as normalized area) are calculated based on the length, width, and area of the bounding box. The shape, texture, color, and size features are combined into a feature vector as the input of the deep learning model.

[0047] A deep learning model based on Convolutional Neural Network (CNN), such as YOLO, Faster R-CNN, or SSD, is selected for foreign object classification.

[0048] Input features: The extracted feature vector is input into the deep learning model and classified through a feature extraction layer (convolutional layer) and a classification layer (fully connected layer).

[0049] Output result: The class probability of each target, representing the confidence that the target belongs to a certain class.

[0050] The classification confidence S is an evaluation of the credibility of the model's detection result, and its value range is [0,1].

[0051] Set a confidence threshold T (such as T = 0.5), and define targets with a classification confidence lower than T as low-confidence targets. Set the total number of detected targets as N and the number of low-confidence targets as Nlow.

[0052] Proportion R of low-confidence targets: ; The higher R is, the more unstable the model classification result or the higher the classification uncertainty. The lower R is, the higher the credibility of the model classification result and the stronger the classification stability.

[0053] After analyzing the proportion of low-confidence targets, a low-confidence target proportion fluctuation index is generated. According to the low-confidence target proportion fluctuation index, the classification stability of the deep learning model algorithm is evaluated. The method for obtaining the low-confidence target proportion fluctuation index is as follows: Collect the time series of the proportion of low-confidence targets ; represents the proportion of low-confidence targets in the T-th frame image (such as in percentage form). T is the total number of frames in the time series. The time series R is segmented and classified into w intervals: ; Denote the boundaries of the \(w\)-th interval. The granularity \(w\) of the interval division can be determined according to the data range and application requirements (for example, 5 intervals, 10 intervals).

[0054] Count the number of data points in each interval , and the expression is: ; Calculate the probability of the \(i\)-th interval: ; is the probability that the proportion of low-confidence occupies the \(i\)-th interval. Calculate the fluctuation index of the proportion of low-confidence targets, and the expression is: ; In the formula, GXH is the fluctuation index of the proportion of low-confidence targets; if the probability of a certain interval , then: .

[0055] When the fluctuation index of the proportion of low-confidence targets is small, it indicates that the time series distribution of the proportion of low-confidence targets is concentrated and changes little, and the classification results of the system are consistent in consecutive detection frames. In other words, the deep learning model has strong robustness and stability in the classification task, can maintain a high classification confidence under different environments and conditions, and has a low risk of false detection and missed detection. This shows that the model has strong reliability in extracting and classifying target features and is suitable for scenarios with high requirements for real-time performance and accuracy.

[0056] When the fluctuation index of the proportion of low-confidence targets is large, it indicates that the time series distribution of the proportion of low-confidence targets is scattered and changes greatly, and the classification results lack consistency in consecutive detection frames. This reflects that there is a large uncertainty in the deep learning model in the classification task, which may be significantly affected by external interferences (such as light changes, background complexity), resulting in an increase in false detection or missed detection phenomena. The classification stability of the model is poor, and its reliability and accuracy need to be improved by optimizing the algorithm, enhancing the training data, or improving the feature extraction method.

[0057] S5: According to the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm, evaluate the accuracy of the image recognition algorithm for track foreign object detection. According to the evaluation results, divide the track foreign object detection results into accurate detection results, incomplete accurate detection results, and inaccurate detection results.

[0058] Convert the foreign object spatio-temporal consistency index and the low-confidence target proportion fluctuation index into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes the prediction of the accuracy value label of the track foreign object detection by the image recognition algorithm for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of the track foreign object detection by all image recognition algorithms as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the accuracy value of the track foreign object detection by the image recognition algorithm according to the model output result, where the machine learning model is a polynomial regression model.

[0059] The method for obtaining the accuracy value of the track foreign object detection by the image recognition algorithm is as follows: Obtain the corresponding function expression from the comprehensive feature vector training data of the trained machine learning model: ; In the formula, is the output function of the model, HAK is the foreign object spatio-temporal consistency index, GXH is the low-confidence target proportion fluctuation index, is the accuracy value of the track foreign object detection by the image recognition algorithm.

[0060] Compare the obtained accuracy value of the track foreign object detection by the image recognition algorithm with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the accuracy value of the track foreign object detection by the image recognition algorithm with the first standard threshold and the second standard threshold respectively; If the accuracy value of the track foreign object detection by the image recognition algorithm is greater than the second standard threshold, it indicates that the accuracy of the track foreign object detection by the image recognition algorithm is high. At this time, generate a high-accuracy detection signal, and classify the track foreign object detection result as an accurate detection result without further processing; If the accuracy value of the track foreign object detection by the image recognition algorithm is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the track foreign object detection by the image recognition algorithm is medium. At this time, generate a medium-accuracy detection signal, and classify the track foreign object detection result as an incomplete accuracy detection result, which may have the risk of false detection or missed detection and requires further verification or review; If the accuracy value of the track foreign object detection by the image recognition algorithm is less than the first standard threshold, it indicates that the accuracy of the track foreign object detection by the image recognition algorithm is low. At this time, generate a low-accuracy detection signal, and classify the track foreign object detection result as an inaccurate detection result. The result may be a false detection or insufficient model performance, and the model needs to be optimized or the algorithm needs to be replaced.

[0061] When the accuracy value of the image recognition algorithm for detecting track foreign objects is lower than the first standard threshold, manifested as large fluctuations in classification results, frequent false detections and missed detections, it indicates that the model's adaptability to feature extraction or classification tasks is insufficient. At this time, the existing model can be optimized to improve its robustness and accuracy by methods such as increasing training data (especially complex scenarios or abnormal samples), improving the model architecture (such as introducing a multi-layer attention mechanism), optimizing hyperparameters (such as learning rate, regularization coefficient), and adopting data augmentation techniques.

[0062] If the existing model still cannot meet the detection requirements after multiple optimizations, or performs poorly in specific scenarios (such as extreme lighting conditions or when there are many dynamic interfering objects), a suitable algorithm needs to be replaced. For example, more advanced object detection models (such as YOLOv8, DETR) can be tried, or a multi-modal algorithm can be introduced to combine images and other sensing data (such as lidar, thermal imaging) for detection to fundamentally improve the detection performance.

[0063] S6: For incomplete accuracy detection results, predict the degree of abnormality of the accuracy of the image recognition algorithm for detecting track foreign objects in the subsequent detection time period. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

[0064] For incomplete accuracy detection results, that is, the accuracy value of the image recognition algorithm for detecting track foreign objects generated within a fixed detection time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. Collect the accuracy values greater than or equal to the first standard threshold and less than or equal to the second standard threshold in the subsequent detection time period, establish a corresponding data set, calculate the mean and standard deviation of the data set, and after analyzing it, predict the degree of abnormality of the accuracy of the image recognition algorithm for detecting track foreign objects in the subsequent detection time period.

[0065] If the mean of the accuracy values in the data set is greater than or equal to the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is less than the reference threshold of the standard deviation of the accuracy values, the detection results are stable and the accuracy is high. At this time, no warning signal is generated, indicating that the system performance is good and no adjustment is required, and the existing detection settings can be maintained.

[0066] If the mean of the accuracy values is greater than or equal to the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is greater than or equal to the reference threshold of the standard deviation of the accuracy values, the detection accuracy fluctuates greatly, but the overall accuracy is acceptable. At this time, a level-three warning signal is generated, indicating that the system detection is not stable enough and may be affected by the external environment (such as lighting changes, interfering objects). The volatility problem should be optimized, such as adjusting algorithm parameters or increasing data.

[0067] If the mean accuracy value is less than the reference threshold of the mean accuracy value, and the standard deviation of the accuracy value is greater than or equal to the reference threshold of the standard deviation of the accuracy value, the detection accuracy is low and the fluctuation is significant. At this time, a first-level warning signal is generated, indicating that the system performance is poor, and improvement measures should be taken, including optimizing the model or improving the image acquisition quality (such as increasing the resolution or frame rate).

[0068] If the mean accuracy value is less than the reference threshold of the mean accuracy value, and the standard deviation of the accuracy value is less than the reference threshold of the standard deviation of the accuracy value, the detection accuracy is low, but the result is relatively stable. At this time, a second-level warning signal is generated, indicating that the model may have insufficient adaptability to specific scenarios, and the model should be optimized or the algorithm should be switched.

[0069] In this embodiment, high-definition cameras are set along the track to obtain track images in real time, and preprocessing such as denoising, contrast enhancement, and color correction is performed on the images. The region of interest in the track area is extracted using an image segmentation algorithm, and object detection is performed based on a convolutional neural network to identify potential track foreign objects. Further, by analyzing the position changes and motion directions of the track foreign objects in consecutive frames, the spatio-temporal consistency of the detection results is verified to determine whether it is a real threat. Multidimensional features such as shape, texture, color, and size are extracted for real threat targets, the classification confidence is output, and the proportion of low-confidence targets is analyzed to evaluate the classification stability of the model. Combining spatio-temporal consistency and classification stability, the accuracy of the image recognition algorithm is evaluated, and the detection results are divided into accurate detection, incomplete accurate detection, and inaccurate detection. For incomplete accurate detection results, the degree of abnormality is predicted by analyzing the mean accuracy and fluctuation of subsequent detections. If the degree of abnormality is high, the image acquisition resolution or frame rate is increased to enhance the clarity of the track image, thereby improving the detection performance and reliability.

[0070] Embodiment 2, please refer to Figure 2 As shown, the intelligent track foreign object detection system based on image recognition in this embodiment includes an image acquisition and preprocessing module, an image segmentation and object detection module, a spatio-temporal consistency analysis module, a classification stability evaluation module, a detection result evaluation and classification module, and a dynamic optimization and abnormality prediction module; Image acquisition and preprocessing module: Real-time acquisition of image data of the track area through high-definition cameras set along the track, and preprocessing the acquired image data, including denoising, contrast enhancement, and color correction; Image segmentation and object detection module: Segment the track area using an image segmentation algorithm, extract the region of interest in the track area, and perform object detection on the extracted region of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; Spatio-temporal Consistency Analysis Module: Obtain the spatio-temporal data of the track foreign object in different frame images, analyze the position changes and movement directions of the track foreign object in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection result, and determine whether the detection target is a real threat; Classification Stability Evaluation Module: After determining that the detection target is a real threat, extract the multi-dimensional features of the detected foreign object, including shape, texture, color and size, output the classification confidence through a deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm; Detection Result Evaluation and Classification Module: Evaluate the accuracy of the image recognition algorithm for track foreign object detection based on the spatio-temporal consistency of the track foreign object detection result and the classification stability of the deep learning model algorithm. According to the evaluation result, divide the track foreign object detection result into accurate detection result, incomplete accurate detection result and inaccurate detection result; Dynamic Optimization and Anomaly Prediction Module: For the incomplete accurate detection result, predict the degree of anomaly in the accuracy of the image recognition algorithm for track foreign object detection in the subsequent detection time period. If the degree of anomaly is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

[0071] As mentioned above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application.

Claims

1. An intelligent detection method for track foreign objects based on image recognition, characterized in that: It includes the following steps: S1: Real-time obtain image data of the track area through high-definition cameras set along the track, and preprocess the obtained image data, including denoising, contrast enhancement, and color correction; S2: Use an image segmentation algorithm to segment the track area, extract the regions of interest within the track area, and perform object detection on the extracted regions of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; S3: Obtain the spatio-temporal data of the track foreign objects in different frame images, analyze the position changes and movement directions of the track foreign objects in consecutive frames, verify the spatio-temporal consistency of the track foreign object detection results, and determine whether the detection target is a real threat; S4: After determining that the detection target is a real threat, extract the multi-dimensional features of the detected foreign object, including shape, texture, color, and size, output the classification confidence through the deep learning model, analyze the proportion of low-confidence targets, and evaluate the classification stability of the deep learning model algorithm; S5: Evaluate the accuracy of the image recognition algorithm for detecting track foreign objects based on the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm. According to the evaluation results, divide the track foreign object detection results into accurate detection results, partially accurate detection results, and inaccurate detection results; S6: For the partially accurate detection results, predict the degree of abnormality in the accuracy of the image recognition algorithm for detecting track foreign objects during the subsequent detection time period. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

2. The intelligent detection method for track foreign objects based on image recognition according to claim 1 is characterized in that: In S2, use an image segmentation algorithm to segment the track area, extract the rail edges and straight structures through edge detection and Hough transformation to distinguish the left and right sides of the track; use a semantic segmentation model to classify each pixel in the image, separate the track area from the background area, and label the pixel tags of the track area for the model to learn the track features; Based on the segmentation result, retain the image part of the track area; Define the region of interest ROI within the track area, delimit the ROI boundary through the geometric characteristics of the track width and spacing; use a convolutional neural network model to perform object detection on the extracted ROI region, select the input track area image and ROI label, combine the labeled foreign object positions and categories for model training, and optimize the model using classification loss and regression loss; output the bounding box coordinates, classification labels, and confidence of the foreign object.

3. The intelligent detection method for track foreign objects based on image recognition according to claim 2, characterized in that: In S3, after analyzing the position changes and movement directions of the track foreign objects in consecutive frames, generate a foreign object spatio-temporal consistency index. The method for obtaining the foreign object spatio-temporal consistency index is: Set the target trajectory sequence as , which represents the position sequence of the detected track foreign object in consecutive frames, and set the reference trajectory sequence , which represents the standard motion mode of the track foreign object; Use the Euclidean distance to calculate the distance between two trajectory points , and the expression is: ; where is the position coordinate of the target trajectory point, is the position coordinate of the reference trajectory point, and construct the cumulative distance matrix D: ; where represents the cumulative minimum distance from the starting point of the trajectory to the point, is the Euclidean distance between the current points, represents the optimal path selection from the previous position to the current point, and the boundary conditions are: ; ; ; In the cumulative distance matrix, find the optimal path P from the starting point (1,1) to the ending point (m,n), and the path consists of a series of points that satisfy: ; represents selecting the path P to minimize the sum of the cumulative distances of all points on the path, and calculate the spatio-temporal consistency index of the foreign object, and the expression is: ; In the formula, L is the length of the optimal path, is the Euclidean distance on the path, and HAK is the spatio-temporal consistency index of the foreign object.

4. The intelligent detection method for track foreign objects based on image recognition according to claim 3, characterized in that: In S4, after analyzing the proportion of low-confidence targets, generate a low-confidence target proportion fluctuation index. The method for obtaining the low-confidence target proportion fluctuation index is: Collect the time series of the proportion of low-confidence targets ; denotes the proportion of low-confidence targets in the T-th frame image, where T is the total number of frames in the time series. The time series R is segmented and classified into w intervals: ; denotes the boundary of the w-th interval; count the number of data points in each interval , and the expression is: ; Calculate the probability of the i-th interval: ; is the probability that the proportion of low confidence falls into the i-th interval. Calculate the fluctuation index of the proportion of low confidence targets, and the expression is: ; In the formula, GXH is the fluctuation index of the proportion of low confidence targets.

5. The intelligent detection method for track foreign objects based on image recognition according to claim 4, characterized in that: In S5, evaluate the accuracy of the image recognition algorithm for detecting track foreign objects based on the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm; Convert the foreign object spatio-temporal consistency index and the low-confidence target proportion fluctuation index into a comprehensive feature vector, and use the comprehensive feature vector as the input of the machine learning model. The machine learning model takes the prediction of the accuracy value label of the image recognition algorithm for detecting track foreign objects for each group of comprehensive feature vectors as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of all image recognition algorithms for detecting track foreign objects as the training target. Train the machine learning model until the sum of the prediction errors reaches convergence and then stop the model training. Determine the accuracy value of the image recognition algorithm for detecting track foreign objects according to the model output result, where the machine learning model is a polynomial regression model.

6. The intelligent detection method for track foreign objects based on image recognition according to claim 5, characterized in that: In S5, according to the evaluation results, divide the track foreign object detection results into accurate detection results, incomplete accurate detection results, and inaccurate detection results, specifically as follows: Compare the obtained accuracy value of the image recognition algorithm for detecting track foreign objects with the gradient standard thresholds. The gradient standard thresholds include a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold. Compare the accuracy value of the image recognition algorithm for detecting track foreign objects with the first standard threshold and the second standard threshold respectively; If the accuracy value of the image recognition algorithm for detecting track foreign objects is greater than the second standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is high. At this time, generate a high-accuracy detection signal and divide the track foreign object detection result into an accurate detection result; If the accuracy value of the image recognition algorithm for detecting track foreign objects is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is medium. At this time, generate a medium-accuracy detection signal and divide the track foreign object detection result into an incomplete accurate detection result; If the accuracy value of the image recognition algorithm for detecting track foreign objects is less than the first standard threshold, it indicates that the accuracy of the image recognition algorithm for detecting track foreign objects is low. At this time, generate a low-accuracy detection signal and divide the track foreign object detection result into an inaccurate detection result.

7. The intelligent detection method for track foreign objects based on image recognition according to claim 1, wherein: In S6, for the incomplete accurate detection results, predict the degree of abnormality of the accuracy of the image recognition algorithm for detecting track foreign objects in the subsequent detection time period, specifically as follows: For the incomplete accurate detection results, that is, the accuracy value of the image recognition algorithm for detecting track foreign objects generated in the fixed detection time period is greater than or equal to the first standard threshold and less than or equal to the second standard threshold. Collect the accuracy values greater than or equal to the first standard threshold and less than or equal to the second standard threshold in the subsequent detection time period, and establish a corresponding data set. Calculate the mean and standard deviation of the data set, and after analyzing it, predict the degree of abnormality of the accuracy of the image recognition algorithm for detecting track foreign objects in the subsequent detection time period according to the analysis result.

8. The intelligent detection method for track foreign objects based on image recognition according to claim 7, characterized in that: If the mean of the accuracy values in the data set is greater than or equal to the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is less than the reference threshold of the standard deviation of the accuracy values, the detection result is stable and the accuracy is high. At this time, no warning signal is generated, no adjustment is required, and the existing detection settings can be maintained; If the mean of the accuracy values is greater than or equal to the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is greater than or equal to the reference threshold of the standard deviation of the accuracy values, the detection accuracy fluctuates greatly. At this time, a third-level warning signal is generated, and the algorithm parameters are adjusted or data is increased; If the mean of the accuracy values is less than the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is greater than or equal to the reference threshold of the standard deviation of the accuracy values, the detection accuracy is low and the fluctuation is significant. At this time, a first-level warning signal is generated, and improvement measures should be taken, including optimizing the model or improving the image acquisition quality; If the mean of the accuracy values is less than the reference threshold of the mean of the accuracy values, and the standard deviation of the accuracy values is less than the reference threshold of the standard deviation of the accuracy values, the detection accuracy is low. At this time, a second-level warning signal is generated, and the model should be optimized or the algorithm should be switched.

9. An intelligent detection system for track foreign objects based on image recognition, which is used to implement an intelligent detection method for track foreign objects based on image recognition according to any one of claims 1-8, characterized in that: Including an image acquisition and preprocessing module, an image segmentation and target detection module, a spatio-temporal consistency analysis module, a classification stability evaluation module, a detection result evaluation and classification module, and a dynamic optimization and anomaly prediction module; Image acquisition and preprocessing module: Real-time acquisition of image data of the track area through high-definition cameras set along the track, and preprocessing of the acquired image data, including denoising, contrast enhancement, and color correction; Image segmentation and target detection module: Using image segmentation algorithms to segment the track area, extracting the regions of interest within the track area, and performing target detection on the extracted regions of interest based on a deep learning model to identify potential track foreign objects. The deep learning model includes a convolutional neural network; Spatio-temporal consistency analysis module: Obtaining the spatio-temporal data of track foreign objects in different frame images, analyzing the position changes and movement directions of track foreign objects in consecutive frames, verifying the spatio-temporal consistency of track foreign object detection results, and determining whether the detection target is a real threat; Classification stability evaluation module: After determining that the detection target is a real threat, extracting multi-dimensional features of the detected foreign objects, including shape, texture, color, and size, outputting the classification confidence through a deep learning model, and analyzing the proportion of low-confidence targets to evaluate the classification stability of the deep learning model algorithm; Detection result evaluation and classification module: Evaluating the accuracy of the image recognition algorithm for detecting track foreign objects based on the spatio-temporal consistency of the track foreign object detection results and the classification stability of the deep learning model algorithm. According to the evaluation results, classifying the track foreign object detection results into accurate detection results, partially accurate detection results, and inaccurate detection results; Dynamic optimization and anomaly prediction module: For partially accurate detection results, predicting the degree of abnormality of the accuracy of the image recognition algorithm for detecting track foreign objects during the subsequent detection time period. If the degree of abnormality is high, increase the resolution or frame rate of image acquisition to enhance the clarity of the track image.

Images