Ship Data Management And Analysis System Based On Double Deep Q-Network

Researchers in marine science have been using traditional manual single operations to detect targets at sea without real-time,which has led to cumbersome processes and an urgent need for an automated platform to manage and analyze sonar data in real-time.In addition,researchers’ initial judgment of target characteristics based on sonar data is affected by noise,resulting in significant errors,and algorithms need to be designed to correct the initial judgment results.The ship data management and analysis system based on Double Deep Q-Network visualize the sonar data as waveforms and greyscale maps and store the collected audio and video data and the processed audio data in local files and databases.This system is composed of the audio and video acquisition and processing module,the database module,and the target feature information prediction module.The audio and video acquisition and processing module call MATLAB algorithms in real-time to process the acquired sonar data and visualize it as waveform and greyscale graphs,while capturing microphone audio data and shipboard host video data;the database module enables data storage and data query;The target features are consist of propeller blade count and rotational speed.The prediction of propeller blade count use a neural network(MLP).The prediction of the target propeller speed adopt the DDQN algorithm to weight the three models of the neural network,random forest,and Lasso regression.Experimental results show that the proposed weighted model obtains better performance in terms of stability and accuracy by comparing with three baselines.This system is based on.NET framework.The Graphical User Interface is implemented by using Win Form controls;for the background processing module,Direct Sound collects audio data and calls FFmpeg to collect video data;the database is SQL Server.In this system,the automation of acquiring,processing,storing,and displaying audio and video data dramatically simplifies the operation of the researchers.In this paper,the target feature prediction algorithm modified the initial results,leading to a much higher accuracy.The system has passed various tests by the A-side and has been officially put into use.