Deep Learning-Based Method for Damage Severity Grading of Black Shell Disease in Hong Kong Oyster (Crassostrea hongkongensis) Caused by Polydora Complex.

The Polydora complex is a globally widespread polychaete parasite known for secreting acidic substances to bore tunnels into oyster shells. This process leads to the formation of dark brown crusts on the inner shell surface, a condition commonly referred to as "black shell disease". Not only does this disease degrade the quality of shellfish, but it also causes substantial mortalities, with the Hong Kong oyster (Crassostrea hongkongensis) being one of its primary hosts. In this study, an improved Deeplabv3+ semantic segmentation model integrated with an attention mechanism was developed to identify and classify the severity of Polydora-induced disease in oysters. First, through systematic data collection and enhancement techniques, a comprehensive dataset of 4590 shellfish disease images was constructed, fully meeting the training requirements of convolutional neural networks (CNNs). Second, transfer learning was employed by leveraging pre-trained model weights, which validated the suitability of the self-constructed dataset for model training. Comparative analysis of multiple segmentation models confirmed Deeplabv3+ as the optimal baseline, achieving a mean Intersection over Union (MioU) of 88.76% and an average precision of 94%. To further enhance performance, the Deeplabv3+ model was upgraded by incorporating the Convolutional Block Attention Module (CBAM), a dual-module mechanism integrating channel and spatial attention. Compared to the original model, the improved version exhibited significant performance gains: MioU increased by 1.19%, average precision rose by 1.78%, and mFscore improved by 0.68%. Finally, an efficient and user-friendly web-based system was designed for grading and identifying Polydora disease severity. This system can accurately segment the diseased regions on the inner oyster shell, calculate their proportion relative to the total inner shell area, and precisely grade the disease severity, thereby providing robust technical support for the monitoring and prevention of oyster black shell disease.