Research On KCF-based Object Tracking For Forward-looking Sonar

As more and more countries attach importance to marine resources,gradually put forward plans for the development of marine resources,but also gave rise to a huge demand for marine resources exploration technology and mining technology,have increased the resources invested in the field of marine exploration.The ocean exploration environment is different from the general gas-mediated environment.water as a transmission medium greatly affects the detection results,visible light and infrared detection technology in the ocean exploration is difficult to cope with the more complex ocean conditions,while underwater sonar equipment with the advantage of long propagation distance in the underwater detection has been widely used,but the sonar image has low resolution,noise and other problems,this paper pin based on these characteristics,proposed a method based on deep learning with the ocean exploration technology.Based on these characteristics,this paper proposes a target detection and tracking method based on deep learning and kernel correlation filtering.The specific research contents are as follows.Firstly,the target detection method of forward-looking sonar images is studied.For the problems of unbalanced size and number of samples in the forward-looking sonar data set and low detection accuracy of small targets,the training data set is expanded by data enhancement of small targets and increasing the number of small targets,so as to improve the detection accuracy of small targets.Secondly,the influence of feature extraction on the target tracking results is studied for the characteristics of forward-looking sonar images.For the general workflow of feature extraction followed by feature fusion,and then sent to the detector to get the tracking results,a decision fusion method based on the peak parametric ratio of the feature output response is proposed,and the reliability of the method is verified through experiments.Again,to address the tracking failure of the forward-looking sonar image target in the case of noise interference,combined with the characteristics of the Kalman filter based on the motion model tracking,by setting the threshold parameters to determine whether the current is suitable for tracking using the kernel correlation filter,the fusion design of the Kalman filter and the kernel correlation filter,using real data to verify the accuracy of the algorithm.Finally,this paper improves the scale adaptive tracking method based on the scale pool.In order to avoid the traversal of the scale pool and to improve the speed and accuracy of the tracking model,we propose a method to filter the scale values by expanding the range of scale values in the scale pool and setting a threshold parameter.