Small-Object Feature-Enhanced Visual Segmentation Model Based on BlendMask for Evaluating Sludge Performance

Anaerobic granular sludge (AnGS) activity measurement has been a hot research topic in the field of wastewater treatment. The traditional method of evaluating activity by biochemical reaction is time-consuming, and in recent years, the method based on image color modeling is faster but less accurate. This article summarizes the shortcomings of the existing measurement methods and proposes an evaluation method based on instance segmentation algorithm and feature association activity. To address the challenge of acquiring high-quality images, a novel image acquisition device and a data production method have been developed to obtain datasets of optimal quality. The instance segmentation method has been improved to address the problem that the proportion of small targets in images is high. Specifically, add attention mechanism, deformable convolutional network (DCN), and augmented feature pyramid network (FPN) to improve the feature extraction ability for small targets. The association of features with active labels for analysis facilitates rapid and precise evaluation. A series of experiments has demonstrated that the enhanced BlendMask instance segmentation model and activity evaluation process exhibit superior segmentation precision and evaluation accuracy in sludge particle scenarios compared with the existing methods.