Improvement of the YOLOv7 model for algae detection in complex backgrounds

Algae in aquatic environments are key indicators of water quality and health monitoring. However, algae detection always faces challenges due to the small size of the targets and the difficulty in feature extraction. In this paper, we propose an improved RSW-YOLO model based on YOLOv7 to improve the detection performance. Firstly, the Receptive-Field Coordinated Attention Convolution (RFCAConv) is used to replace a portion of convolutions, allowing the model to focus on the most important features and enhance feature fusion. Secondly, a Series-Parallel Spatial Pyramid Pooling (SPSPP) module is designed to improve the model’s feature representation capabilities. Finally, the WIoU v3 boundary loss function is employed to dynamically adjust loss weights, improving detection accuracy. The experimental results indicated that the improved RSW-YOLO outperforms the baseline model, achieving 2% increase in [email protected] and a 0.3% improvement in recall (R). This improved algorithm provides valuable insights for algae detection.