Research Of Intelligent Full-ocean-depth Fishing Device And Target Detection Algorithm

The scientific community defines the area between 6000 meters and 11000 meters(full-ocean-depth)as hadal.As a vertebrate,hadal fish is located at the top of the hadal bio-chain and plays an important role in the study of hadal science,hadal biology and hadal ecology.Focusing on the needs of hadal fish catching,this thesis takes the full-ocean-depth intelligent fishing device as the research object,and aims to realize the precision fishing in the full-ocean-depth.With the hadal fish target detection as the core,the thesis focuses on the following aspects:(1)Based on the development of intelligent full-ocean-depth fishing device,the requirement analysis is carried out.The device is proposed to have four functions: deep-sea landing,target biological intelligent detection and capture,deep-sea environmental information acquisition and device sea surface recovery.The syste m structure design of the intelligent full-ocean-depth fishing device is then carried out.In which the carrier subsystem realizes the function of deep-sea landing to achieve the perform of deep-sea operation;the central management subsystem and sampling subsystem realize the function of deep-sea biological target detection and automatic capture,deep-sea environmental information acquisition,and realize the closed-loop control of the capture process to achieve precise intelligent fishing;the common-load subsystem ensure the device to be found at the surface of the sea immediately.(2)This thesis designs a fishing cage which is one of the core mechanical structures of the intelligent full-ocean-depth fishing device.The fishing cage adopts double open door sliding track structure.One of the reasons is that the connecting rod driven b y torsion spring drives the cage door to close along the sliding track to ensure the stable operation of the cage door.The other reason is to reduce the moving distance of the single cage door and to increase the closing speed of the cage door.The cage door is automatically controlled by single-chip computer and electromagnetic switch.In detail,the single-chip computer receives the target detection information and makes analysis and judgment.After determining the detected target,the control signal is output to the electromagnetic switch to control the closure of the left and right cage door,which realizes the intelligent and automation of the device.(3)Aiming at the target bio-intelligent detection of the hadal fish,the design and algorithm of target detection module are studied.This thesis describes the research and development of traditional target detection algorithm and deep learning target detection algorithm.An algorithm based on OpenCV and combining motion detection with Haar feature detection,Motion-Haar algorithm,is designed in this thesis.Its purpose is to use the motion detection method to determine whether there is a moving target in the image,and to detect the hadal fish target in the image through the Haar feature detection algorithm,so as to improve the real-time detection and reduce energy consumption.Furthermore,base on the implementation effect of the Motion-Haar algorithm,an Deep-YOLO algorithm,improved YOLOv3-Tiny algorithm,which integrates image pre-processing and target ranging functions,has been designed.Its main innovation is to enhance detection performance through network deepening.Median filter is introduced to remove noise interference and increase target ranging function to avoid false alarm beyond the region,which improve the detection accuracy.(4)Sample extraction and algorithm testing.Picture preparation,interception and screening of hadal fish were carried out.The training sample set was made by Haar feature detection process and YOLOv3-Tiny detection process,meanwhile,the target detection classifier and target detection network were trained respectively.The performance index of target detection is determined.Motion-Haar algorithm and Deep-YOLO algorithm are implemented on Windows 10 platform.Video detection is used to test and analyse its real-time performance,accuracy and false alarm rate.The test and analysis results show that the Motion detection in the motion-haar algorithm can effectively detect whether there is a moving target in the image,but the Haar feature detection cannot detect the hadal fish target.Deep-YOLO algorithm can detect hadal fish targets in the image in real time,with high real-time performance,accuracy and low false alarm rate,but there is a little overfitting phenomenon.Deep-YOLO algorithm is very suitable for the detection and capture needs of deep-sea fish such as lionfish and mousetail,providing technical support for complete fish capture and scientific research and analysis of deep-sea fish.