The inherent characteristics of algae and disturbances in underwater environments make it difficult to extract effective feature information from algal images, leading to suboptimal performance of traditional object detection models and frequent missed detections. To address these challenges, this study constructs a semantic-augmented feature fusion framework (AlgaeFusion-YOLO) based on the YOLO11 backbone network. It innovatively introduces the WTConv convolutional module to enhance perception of algal morphology and texture, and employs the CBAM dual-channel attention mechanism to adaptively focus on information-rich regions, thereby strengthening target information perception. The BiFPN skip-connection mechanism facilitates information transfer between feature maps at different scales, improving the extraction of algae features from microscopic to macroscopic levels. To address the category imbalance between rare and common algae genera, few-shot learning and CLIP semantic supervision techniques are introduced, effectively resolving the accuracy gap in detecting scarce algae species. Experimental results demonstrated that the proposed model achieved an mAP of 83.9% and mAP@50–95 of 56.4% on the algal dataset, representing improvements of 11.5% and 14.3% over the YOLO11 baseline framework. This fully demonstrates the model's adaptability and precision in algal detection within aquatic scenes.
Wu et al. (Wed,) studied this question.