Nile Tilapia Pathogens Detection Using CNN Model Infused with Efficient Channel Attention Mechanism

Résumé fourni par la source

Nile Tilapia, a highly resilient fish species crucial to aquaculture, faces significant health threats from various infectious diseases. These diseases are caused by bacterial, viral, and parasitic pathogens that affect fish health, productivity, and overall sustainability, as the Food and Agriculture Organization (FAO) reported. Parasitic infections from monogeneans, streptococcosis from Streptococcus agalactiae, columnaris disease from Flavobacterium columnare, Tilapia Lake Virus (TiLV), and Motile Aeromonad Septicemia (MAS) from Aeromonas species and so on. Effective disease treatment in aquaculture depends on prompt and accurate identification due to the overlapping symptoms of these infections. Though reliable, traditional diagnostic methods, like PCR and histology, can be time-consuming and require specialized equipment and expertise, limiting their accessibility. In this study, we used several pre-trained deep learning models to address the identification challenge, including VGG16, MobileNetV2, DenseNet201, ResNet101, and ResNet152 for disease detection in Nile Tilapia. Additionally, we proposed a customized CNN model enhanced with an Efficient Channel Attention (ECA) mechanism to focus on the most relevant features in the input data. Our proposed model demonstrated strong generalization capabilities, achieving a training accuracy of 99.85%, validation accuracy of 93.44%, and 96.86% test accuracy with precision 96.08%, recall 95.8%, and f1-score of 95.88% on the Enhancing Disease Detection in Nile Tilapia (EDDNT) dataset. For the ablation study we have employed Convolutional Block Attention Module (CBAM), Squeeze and Excitation Module (SE-Net) with CNN model and Dense layers.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.