Research On Machine Vision-based Environment Awareness Technology For Unmanned Surface Vehicles

In recent years,with the rapid growth of the world economy and the rapid development of science and technology,more and more countries and regions have focused their development on the vast and abundant ocean.In order to take the lead in the fields of resource development and marine defense,countries around the world have vigorously carried out research and development of marine equipment.Unmanned Surface Vehicles(USV)is an integral part of advanced marine equipment.In fact,both in the military and civilian fields,surface unmanned boats have extremely important value.Accelerating the research of unmanned boat-related technologies has extremely important strategic significance for China to improve the level of marine equipment.This article relies on the "huster-68" unmanned boat developed by Huazhong University of Science and Technology and its supporting scientific research platform,focusing on the research of unmanned boat environment perception technology based on optical sensors.Text research directions include: low-contrast water surface image enhancement technology,water surface target detection and recognition model compression technology and water surface target tracking technology.The details are as follows:When enhancing low-contrast water surface images,traditional image enhancement algorithms need to manually set parameters,and the operating efficiency is low.In response to the above problems,this paper reconstructs the MSR algorithm with a convolutional neural network and proposes the MSRN model.On this basis,the channel-level visual attention mechanism and codec structure are introduced to improve the structure of the MSRN network,and the MSSEN model is proposed.Experiments show that the MSSEN model has obvious enhancement effect on low-contrast water surface images.In addition,based on the MSSEN network,this paper proposes a framework that combines low-level visual tasks and high-level visual tasks to complete image enhancement and surface target detection and recognition in an end-to-end manner.The task of improving the performance and efficiency of the entire algorithm.When deploying a surface target detection and recognition model on an unmanned boat,the traditional detection and recognition model is not accurate enough,and the detection and recognition model based on deep learning has a large number of parameters and cannot be run on embedded devices.In view of the above problems,this paper first introduces the YOLO series of algorithms into the field of surface target detection and recognition,and then proposes four types of convolution kernel parameter evaluation indicators from the perspective of weight activation measurement and parameter optimization measurement,and proposes a single-stage dynamic weighted pruning strategy Pruning compression of YOLO series models.Experiments show that the compressed model in this paper significantly reduces the amount of parameters and energy consumption while keeping the detection and recognition accuracy basically unchanged,and the inference speed is greatly improved.When tracking water surface targets,the template matching algorithm and the common correlation filtering algorithm cannot cope with the changing conditions of target shape,scale and color,etc.,and the tracking algorithm based on deep learning has low calculation efficiency and cannot meet Engineering needs.In view of the above problems,this paper first studies the water surface target tracking algorithm based on kernel correlation filtering and analyzes the shortcomings of its sampling method.On this basis,the water surface target tracking algorithm based on background perception related filtering is studied,which improves the sampling method.Mentions tracker performance.Experiments show that the water surface target tracking algorithm based on background perception correlation filtering still performs well under the harsh conditions such as complex background and unclear target.