YOLO-CBNet: A robust attention-enhanced detection framework for underwater fish recognition in aquaculture environments.

Abstract

The accurate identification of fish species remains a pivotal challenge in aquaculture, as it directly influences population management, health monitoring, and overall production efficiency. Yet, underwater environments often present difficult conditions such as reduced visibility, suspended particles, and uneven lighting that significantly limit the performance of traditional recognition approaches. In this study, we introduce YOLO-CBNet, an enhanced architecture designed to overcome these constraints and provide more dependable monitoring in real-world aquaculture systems. The method adopts a two-stage strategy. First, a Contrast Limited Adaptive Histogram Equalization (CLAHE) module is employed to counter low contrast, turbidity, and color distortion, restoring essential visual information. Second, the YOLOv11 framework is refined through the integration of the Convolutional Block Attention Module (CBAM) and a Bidirectional Feature Pyramid Network (BiFPN), a combination intended to improve multiscale feature fusion and reinforce the detection of small or partially occluded fish. Experimental results indicate that YOLO-CBNet delivers a clear performance boost, the precision score reaches 0.951, surpassing the baseline YOLOv11 model, which achieved 0.911 under the same conditions. These improvements highlight YOLO-CBNet's potential as a reliable solution for underwater detection and recognition tasks within modern aquaculture environments.