Attention-Based Mechanism for Fish Disease Classification in Aquaculture

Fish is a rich supply of proteins. Due to its high demand, aquaculture has been growing steadily. However, the activity is vulnerable to many diseases. Recent developments in computer vision and internet of things have enabled the automation of aquaculture operations. The lack of public fish disease dataset and complex underwater environments have limited the advancement of automatic fish disease detection and classification. This study proposes attention-based mechanism with background removal for fish disease classification. We focus on using strongly discriminative features of the infected fish regions and weakening regions of low interest using convolutional block attention module (CBAM), which is added to the pretrained classification models to sequentially infer attention maps along the channel and spatial dimensions for every intermediate feature map. The attention maps are then multiplied to the input feature map for adaptive feature refinement. The models are trained, validated and tested on a custom dataset with image samples collected from various internet sources. The performance of the attention-based models is compared with the baseline. The results indicate that ResNet50 with CBAM achieves 89.9% of accuracy, precision of 89.9%, recall of 89.3% and 89.7% of F1-score. Conclusively, attention mechanism improves fish disease classification performance.