A Method for Floc Particle Recognition in the Flocculation and Sedimentation Process Based on the Improved ECA-YOLOv8 Model

Flocculation and sedimentation processes are crucial purification steps in water treatment, where the dosing conditions play a vital role in flocculation efficiency and effluent water quality. Numerous studies have shown that floc particles are important parameters for characterizing the dosing treatment in the flocculation and sedimentation process. However, floc particles exhibit significant changes in shape and size during flocculation, as they continuously grow, aggregate, and break apart. Additionally, suspended solids in water, lighting conditions, and camera equipment can interfere with the recognition of floc particles in images, making it difficult for traditional methods to achieve accurate identification. To address this challenge, an improved ECA-YOLOv8 model is proposed for floc particle recognition in the flocculation and sedimentation process. In the Backbone, the efficient channel attention (ECA) mechanism is utilized to optimize the model structure, enabling the model to identify features more relevant to floc particles. Meanwhile, the Bidirectional Feature Pyramid Network (BiFPN) is employed to improve the connection method in the Neck part of the YOLOv8 model, allowing for more effective extraction and fusion of features from small particles, thereby enhancing the model's detection performance for small floc particles. Performance comparisons of several different models on a floc particle dataset demonstrate that the proposed method achieves a 1.9% increase in F1 score and a 2.1% increase in mAP@50, effectively recognizing floc particles.