Research On Intrusion Detection Of Railway Track Foreign Objects Based On Image Processing

The safety of railway operations is an increasing concern due to the rapid development of China’s railway construction and the constant expansion of its operational mileage.Negligence can cause major financial losses and perhaps put lives in danger if foreign objects aren’t found in time.Thus,it is essential to quickly and accurately detect any entering alien objects.However,because trains are moving faster,there are greater demands on detecting speed and accuracy.In order to conduct research on monitoring-based foreign object intrusion detection on railway tracks,this paper uses the railway perimeter intrusion alarm system.Image segmentation and the YOLOv5 algorithm in deep learning are combined to achieve foreign object target detection on railway tracks and improve detection accuracy and efficiency.The following are the thesis’ s primary research components.Firstly,a railway track intrusion foreign object sample database with more than 6000 images was first created by performing a requirement analysis for railway perimeter security,building a railway track foreign object intrusion database,collecting various railway scene images as well as foreign object intrusion images,and enhancing the data by local cropping and changing the chromaticity brightness.This dataset can be used to train and evaluate the enhanced YOLOv5 model.Secondly,by examining the railway scenes,we discover that they are characterized by a fixed composition of elements,areas that are primarily separated by straight lines,and railway ends that radiate outward in all directions from the center.As a result,the Hough transform is used to determine the image’s feature maxima,and the Gaussian convolution kernel is then rotated in accordance with the maxima to specifically improve the linear border point weights.The boundary point weights are produced by merging the Gaussian convolution operation with the pixel coordinates in various channels with the addition of the feature channels through the color model transformation.In order to recognize track regions,the fragmented regions are finally combined in a continuous cycle through threshold screening to produce local regions of the railway landscape.These local regions are then fed into the improved YOLOv5 model.Thirdly,the CLAHE(Contrast Limited Adaptive Histogram Equalization)algorithm is added to the YOLOv5 network structure to improve the contrast of the image and lessen the impact of light and camera shake on the results of detection;this is done to address the issue of unclear images in railway scenes and the presence of small targets and obscured targets that are not detected;In order to improve feature extraction and address the issue of obscured targets and small targets being easily missed,the convolutional block attention model is introduced on the foundation of the original YOLOv5 algorithm.Additionally,the GIo U loss function is replaced by the -CIo U loss function as the bounding box regression loss function to speed up the model’s convergence and enhance the precision of border localization;In order to minimize the number of operations to 1/8 of the original one,the network divides the three-dimensional convolutional operations into 3 × 3 depth convolution and 1 × 1 point convolution.According to the simulation results,the average detection accuracy of m AP(mean Average Precision)can reach 94.1%,and the FPS(Frames Per Second)can reach41 frames per second,which can quickly and precisely detect foreign objects on the railway and completely meet the requirements of real-time target detection.Lastly,the enhanced YOLOv5 method in this work was compared and assessed with SSD,Faster R-CNN,YOLOv3 and YOLOv5 algorithms on the same railway dataset.The improved algorithms in this paper were exposed to ablation experiments to test the unique effect of each modification independently.The findings demonstrate that the suggested technology can fulfill the actual needs of railway field use and is superior and highly accurate when used for foreign body infiltration detection on railway tracks.