Marine Fish Species Classification Using Transfer Learning and Residual Network

Sea fish is one source of food that can be found in coastal waters. Fishermen meet market demand and deliver products to local markets. Fish species are selected using visual observation. Artificial intelligence is expected to help this process by using deep learning. The algorithm that can be used is the Convolutional neural network. Classification of fish species: Research the shape and color of fish in depth so that the machine can help distinguish the type of fish caught. This research used nine classes of marine fish for consumption, including Tongkol, tuna, Kembung, Tenggiri, Kakap, Cakalang, Salmon, Sarden, and Baronang. Each of the nine classes of fish contains 60 training images, including 45 validation images and 10 test images, so the total images used are 534. The architecture used is a custom residual layer with data augmentation. Data limitations handled by augmentation techniques include scaling, rotation, sliding, and reflection. From the results of the research trial scenarios, it can be seen that this research can classify species with an average accuracy of 99.26%, precision of 0.9926, recall of 0.9927, and F-1 score of 0.9927 with an average computational learning time using layers custom about 3 minutes.