Research On Information Acquisition And Object Detection System Towards Offshore Edge Computing

According to data released by the International Maritime Organization(IMO),human factors such as crew fatigue and decision-making errors are the leading causes of maritime accidents,accounting for approximately 60% of all incidents at sea.Adverse weather conditions are the second leading cause,contributing to about 15% of maritime accidents.Providing timely navigational weather information and environmental awareness capabilities for ships at sea can greatly assist in reducing maritime accidents.However,existing systems such as AIS and VDES only provide environmental information between individual vessels and do not offer comprehensive environmental data or awareness of the entire navigational route.Therefore,this paper presents the design and implementation of an information collection and target detection system that provides navigational environmental information and awareness capabilities for maritime vessels.Firstly,this study explores an overall design scheme consisting of environmental information collection sub-nodes,video information collection sub-nodes,and an intelligent buoy edge computing master node.Based on this scheme,a hardware experimental platform is designed and constructed.Secondly,in order to ensure efficient and reliable transmission of data collected by multiple environmental information collection sub-nodes to the intelligent buoy edge computing master node,and considering the types of collectable environmental information at sea as well as the task transmission requirements between environmental information subnodes and the intelligent buoy edge computing master node,this paper proposes and implements an environmental information collection frame structure for gathering data from multiple environmental information collection sub-nodes.Furthermore,to enhance the ship’s environmental awareness capability,this paper presents a deep neural network model called Tiny Hourglass(TH)based on keypoint information detection.Building upon the TH deep neural network model,an improved object detection algorithm named YOLOv5Lite-TH is proposed.Experimental results demonstrate that the proposed algorithm achieves higher accuracy and mean average precision.Finally,to ensure compatibility and efficient and reliable operation of all system components,the Linux operating system is utilized as the operating system for the intelligent buoy edge computing master node.For the environmental information collection functionality,the Lo Ra module and environmental information collection protocol are employed for development.As for the object detection capability,a UDP-based object detection server is designed and implemented to achieve real-time and efficient detection of target objects from video information collected by the video information collection sub-nodes.Through experimental verification,the system is capable of collecting environmental information from multiple environmental information collection sub-nodes and detecting target vessels from video information collected by video information collection sub-nodes.Therefore,the system can complement AIS or VDES by providing navigational weather,hydrological information,and environmental awareness capabilities for ships navigating at sea.