Research On Underwater Biological Detection Technology Based On Improved Yolov5s Algorithm

Underwater target detection technology can effectively detect and identify common organisms in shallow water(sea cucumber,scallop,starfish,sea urchin),providing important information support for marine research,search and rescue,safety monitoring,marine resource development,and ecological protection.However,at present,underwater target detection still faces some difficulties,such as: difficulty in underwater data collection;Loss of features caused by factors such as light,underwater sand,and mud;The complexity of underwater physical environment leads to low efficiency and accuracy of underwater detection.In order to achieve real-time and high-precision underwater target detection,this article has carried out research on underwater biological detection technology,the main research contents include the following:(1)In order to solve the problem that underwater target detection algorithms are far less accurate than land detection due to the impact of light changes and sediment,this paper proposes a YOLOv5s-SPP3 underwater high-precision detection algorithm embedded with convolutional block attention mechanism.Combining attention mechanism and pyramid pooling layer,important feature information of underwater images is extracted and fused,effectively reducing false detection and missed detection caused by feature information loss.The traditional non maximum suppression is replaced by center distance non maximum suppression,which optimizes the detection and positioning of overlapping targets and improves the positioning accuracy of the algorithm model.(2)Due to the large amount of computation and many parameters in the underwater target detection network based on YOLOv5 s,the processing time is slow.In order to improve its reasoning speed and reduce the hardware conditions for deployment,a lightweight underwater target detection network based on YOLOv5 s algorithm is constructed.Firstly,by introducing Ghost Conv to form a Ghost Bottleneck module and replacing the Bottleneck in the C3 module to form a C3 Ghost module,the C3 and Conv modules in the network are replaced with C3 Ghost and Ghost Conv to reduce the amount of network parameters and floating point calculations;Secondly,the EAMC3 combination module is added to the last layer of the backbone network to deepen the depth and ability of the feature extraction network.The data obtained after the experiment shows that the amount of network parameters and floating point computation after lightweight are reduced by 40.28% and 46.06%respectively compared to YOLOv5 s,and the FPS reaches 118.13,and the detection accuracy increases by 1.73%.(3)In order to apply the proposed underwater target detection technology to engineering practice,this study deployed the proposed lightweight network on the web side using the Flask environment and Hypertext Markup Language.Firstly,a web interface containing registration,login,and function selection was written using Hypertext Markup Language;Secondly,by installing the Flask library in the Py Charm engineering environment,instantiating the Flask application and defining some Flask routing to switch between registration,login,and function selection pages,sending requests and responses through the browser to achieve real-time display of webpage detection results,and saving files in the backend.After the above technical research,the lightweight underwater object detection system ultimately designed and implemented in this project can achieve high-precision real-time object detection,and has completed webpage deployment.The research,design,and system implementation of underwater biological detection technology have been completed,with broad application prospects.