MUHAMMAD FIKRI HAEKAL/152018085 Penulis 0411038512 - Galih Ashari Rakhmat, S.Si., M.T. Dosen Pembimbing 1

Fish contains many nutrients such as high protein, it has many benefits for the human body. Many people consume fish as a daily food. However, the majority of people when buying fish have serious concern in choosing fish related to the freshness. Computer vision applications can be used to correctly classify fish freshness. In this research, it was carried out as an initial step in determining the classification model used in the application that can correct the freshness of the fish. The research was conducted by classifying 24 fish freshness classes based on fish eyes by implementing the MobileNetV3 architecture as an extension of MobileNetV1 and MobileNetV2. Squeeze-And-Excitation in MobileNetV3 Architecture has the effect of improving accuracy and reducing the number of parameters. Tests were conducted with several model variations based on different optimizers, epoch, and batch sizes. Based on the results, the model with the best performance is resulted by the hyperparameter such as batch size for training data with 10 batches, and validation data with 10 batches. Optimizer that is being used in this research is ADAM, and number of epochs 100 with MobileNetV3-Small architecture model. Based on the results of the model’s performance evaluation, 68% accuracy, 69% precision, 67% recall and 68% f1-score is resulted using 876 test data with 24 types of fish and various freshness. However, it can be said that MobileNetV3 architecture is the state-of-the-art of fish freshness based on fish eyes.