Accurate Fish Detection under Marine Background Noise Based on the Retinex Enhancement Algorithm and CNN

Underwater detection equipment with fish technology has broad application prospects in marine fishery resources exploration and conservation. In this paper, we establish a multi-scale retinex enhancement algorithm feature-based model to improve underwater accuracy ensure real-time performance. During image preprocessing, the combines bionic structure of retina classical theory filter out environmental noise. The focuses on improving performance small-size targets using deep learning method based convolutional neural network. We compare our current mainstream models (Faster R-CNN, RetinaNet, YOLO, SSDetc.), proposed achieves better performance, mean Average Precision (mAP) 78.31% Miss Rate (mMR) 54.11% open data set. test results for from field experiment prove feasibility stability model.