Answers are grounded in this paper and cite their sources.
Underwater fish disease detection is a vital visual perception task for health monitoring, disease early warning and precision breeding management in aquaculture. In complicated underwater environments, fish lesions suffer from faint boundaries and tiny sizes, while cross-scale fused features of detection models are contaminated by mixed degradation interference. To tackle the above issues, this paper proposes a YOLOv12-based detection method for fish lesions under complex underwater conditions. Firstly, an underwater spatial context receptive field module named CRB is designed. It expands the receptive range of shallow features through two-stage depthwise separable dilated convolution, establishes spatial correlations between local lesion responses and surrounding neighborhood structures and enhances the local structural representation of lesions with faint boundaries and small scales. Secondly, an underwater degradation-aware channel enhancement module DCE is presented. Degradation priors are constructed from image intensity, gradient and local contrast. Combined with mixed pooling to generate channel weights, the module recalibrates lesion-related responses and degradation interference responses after cross-scale feature fusion, so as to suppress the propagation of degradation noise within fused features. Lastly, the experiments are conducted on a self-built underwater largemouth bass lesion detection dataset. The experimental results show that the proposed model achieves Precision of 87.4%, Recall of 81.3%, F1-score of 84.2% and mAP@50 of 85.8%, which outperforms the baseline model by 3.3, 3.7, 3.5 and 3.6 percentage points, respectively. The proposed method serve as an effective visual perception solution for fish lesion detection in complex underwater environments.
No takes yet. Share an insight, caveat, or question.
Jiang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: