Sonar small object detection algorithm based on background suppression and improved multi-scale LSD
To solve the problem of high false alarm and false detection in sonar small object detection under complex underwater environment and low signal to noise ratio,a detection algorithm based on background suppression and improved line segment detection(LSD)is proposed,in which,a sequence of fragments is extracted from original sonar data to construct a multi-period cumulative history image and highlight the moving object trajectory line features;an edge filter operator is designed to effectively filter out part of the background noise,and by combining projection transformation to enhance line features,not only the broken line reconnection is realized,but the residual noise is suppressed a well;Then,the multi-scale LSD algorithm is improved based on the image pyramid to effectively alleviates the over-detection and increases the average length of the line.Finally,the post-processing module is designed by using the time and space consistency of the motion trajectory to merge redundant detection information,and the location accuracy is improved.By the qualitative and quantitative analysis of the lake and sea test data and the visual results of remotely operated vehicle(ROV),divers and hollow ball objects,the false detection rate and lost detection rate of the improved LSD algorithm are reduced by 11.2 percentages and 3.9 percentages respectively,and the error of location falls by 1.495 pixels,compared with traditional LSD algorithm.The results show that the proposed algorithm greatly improves the detection accuracy of sonar small object,which lays an important foundation for subsequent underwater object recognition and tracking.
sonar small object detectionbackground suppressionmulti-scale line segment detection(LSD)sonar image
万方数据
维普
