Automated region split algorithm with pre-trained CNN model for tuna classification

Tuna is an essential fish due to its nutritional content and is one of the most commonly traded fish in the world. These fish are manually separated into various species in the export industry. An automated approach is proposed in this study to speed up these sectors in order to meet the growing demand for tuna and its products. Three species of tuna is used for the study. Initially the fish images are segmented using a pretrained U-Net model fine tuned on custom dataset. The segmented images are applied to a region split algorithm for spliting fish image into three region images. Each region image is applied to three different pretrained VGG16 model. The predictions from the three models are combined by sum rule to get the image wise prediction. Result shows that the proposed system gives an accuracy of 87.54% in classifying tuna into its species types.