Improved YOLO Algorithm Research And Its Application In Underwater Biometrics

In recent years,marine resources have become a key resource for exploitation and utilization.How to accurately identify underwater organisms has become more and more important.In the past,safety could not be guaranteed when sending personnel underwater to shoot and capture for biometric identification.In recent years,the development and application of underwater robots have solved this problem,and the strength of underwater biological object recognition ability is especially important for underwater robots.This paper addresses the key technologies of underwater image blurring as well as object detection in underwater biometric recognition.The research contents are as follows.(1)Underwater image enhancementAccording to the theory of underwater optical imaging,we understand the principle of the scattering phenomenon of underwater light and the absorption effect of seawater itself on the light.The enhancement of underwater images effectively solves the problem of too many blue-green waves in underwater images,while partially restoring the real situation of underwater images and providing a more distinctive image for underwater biometric identification below.(2)Research on the traditional object detection taskThe processed images are put into the classical deep learning model for comparative study,and firstly SSD,as well as YOLO,are put into the experiment.The experimental results show that YOLO has more than 1 percentage point higher accuracy than the SSD model,so YOLO is used as the basis for the subsequent improved model.(3)Improving the YOLO modelSince the YOLO algorithm has the problem of faster object detection but lower recognition accuracy,this paper proposes an improved one by adding a multi-scale feature extraction structure to the neck layer network and a double-head design in the head network.The results of the study show that the improved YOLO model has greatly improved the ability of underwater biometric recognition.