Jellyfish detection algorithm of underwater image based on TBFRNet and reinforcement learning

This paper presents a novel deep learning-based algorithm for jellyfish target detection in underwater images, achieving recognition of 12 jellyfish species using a self-constructed dataset. Compared to terrestrial and other marine organisms, jellyfish exhibit more complex morphological features in optical imagery, posing significant challenges for detection. To overcome this difficulty, this paper initially proposes a target detection backbone network, TBFRNet, which is based on the Transformer architecture in combination with independent component analysis to enhance the accuracy of target detection at the macroscopic level. Subsequently, we design a jellyfish feature reinforcement learning detection module that aligns with human perception to improve the jellyfish feature detection outcome. A lightweight submodule is also designed to reduce computational complexity without sacrificing performance. Experiments on our jellyfish dataset demonstrate that the proposed algorithm outperforms state-of-the-art detection methods in robustness and accuracy, particularly in handling optically diverse marine environments. This work advances the application of optical image processing in marine biology and underwater surveillance.