Lightweight Monitoring Method for Fish Recognition Based on Image Enhancement

A lightweight fish recognition and tracking method was proposed to cope with image degradation, turbidity, and environmental complexity in small river basins, with consideration of the real-time and high-precision requirements of ecological monitoring. An experimental platform was established under various water quality and illumination conditions to simulate four representative scenarios, including clear and turbid water as well as day and night environments, and image and video samples of common fish species such as crucian carp, carp, and grass carp were collected to construct a multi-scene dataset. The lightweight underwater image enhancement algorithm FA+Net was applied for color correction and detail restoration to improve brightness and contrast, while an improved YOLOv11n-DCGA detection model was developed by introducing the DualConv structure to reduce model parameters and incorporating the C3k2_GSA attention mechanism to strengthen feature extraction. Combined with the ByteTrack multi-object tracking algorithm, stable identification and trajectory tracking of fish targets were achieved. Experimental results showed that FA+Net improved SSIM, UIQM, and UCIQE by 19.81%, 17.69%, and 21.87% compared with traditional algorithms; the YOLOv11n-DCGA model reduced parameters and computational cost by 7.3% and 7.8% while improving mAP0.5, mAP0.75, and mAP0.5: 0.95 by 1.0%, 2.4%, and 1.6%, respectively; after integrating ByteTrack, stable multi-fish recognition and continuous tracking were maintained even under overlapping, fast-moving, and occluded conditions, achieving a detection rate of 43 FPS. The proposed method significantly enhances the accuracy and stability of underwater fish recognition while maintaining real-time performance, providing a feasible technical solution for real-time ecological monitoring system deployment and offering valuable support for fish resource assessment and aquatic ecosystem protection.