Intrusion Object Detection Under Typical Adverse Weather Conditions For High-Speed Railway

Safety is the permanent goal and first priority of high-speed train operation.However,high-speed trains operate in a complex open perimeter environment,facing illegal intrution and illegal construction which bring huge challenges to the safety of high-speed train operation.It has become an inevitable trend to guarantee perimeter safety,improve the monitoring technology of high-speed railway,and promote the intelligent and high-precision development of video surveillance technology along the high-speed railway.The rapid development of video surveillance technology provides a basic guarantee for the safety of the high-speed railway perimeter,however,the existing daytime intelligent video image analysis technologies in the high-speed railway perimeter are affected by typical adverse weather conditions(haze,rain)and large-scale small intrution objects,causing the false positives and false negatives to increase significantly.Therefore,how to improve intrution object detection accuracy under adverse weather conditions is the focus of this thesis.(1)In the hazy image enhancement research,in order to improve the intrusion detection accuracy of intelligent recognition algorithm under hazy condition for highspeed railway perimeter,an image dehazing method based on improved residual block and multi-sampling fusion is proposed.The feature extraction network and the combined loss function are verified and compared through various experiments,which proves that our method achieves the best dehazing results in the evaluation index.(2)In the rain image enhancement research,in order to improve the intrusion detection accuracy of intelligent recognition algorithm under rainy condition for highspeed railway perimeter,a cross-scale fusion-based and fully convolutional-based image rain removal method is proposed.A cross-scale fusion and fully convolutional image rain removal network,combined loss function,and method acceleration experiment are designed according to the distribution characteristics of rain.Evaluation on rain railway dataset proves that the rain removal algorithm has achieved good results.(3)In the research on the detection method of small intrution object in the railway perimeter,in order to improve the intrution detection accuracy of small object,a detection method for small object based on a prior bounding box clustering and attention mechanism is proposed,and the YOLO-SE network structure is designed,the prior anchor cluster and data augmentation are leveraged.Finally,the experiment proves that the YOLO-SE object detection method is better for small object detection in the highspeed railway.(4)Research on the railway perimeter intrusion object detection under adverse weather conditions,On the basis of the previous chapters,the analysis and processing flow of intrusion detection under typical adverse weather conditions for the perimeter of high-speed railway is proposed.For high-speed railway perimeter intrusion detection under hazy conditions,an integrated intrusion detection mehtod under hazy conditions is proposed;for high-speed railway perimeter intrusion detection under rain conditions,an integrated intrusion detection mehtod under rainy conditions is proposed algorithm.In summary,this thesis has conducted an in-depth research on the intrution object detection for railway perimeter under typical adverse weather conditions,starting from the perspectives of railway hazy image enhancement,railway rainy image enhancement,and railway perimeter small intrution object detection.The accuracy of object detection on railway perimeter under typical adverse weather conditions is improved.The research results obtained in a certain sense enrich the research content of intrusion object detection in the railway perimeter under adverse weather conditions at home and abroad,and will also provide theoretical support for managers,practitioners and scholars of high-speed railway perimeter security.