Fish Image Classification Based on MobileNetV2 with Transfer Learning Technique for Robotic Application in Aquaculture

Abstract

Intelligent aquaculture helps solve the problems of traditional aquaculture by applying smart production methods. The combination of technology and the art of farming has opened the door to the smart seafood industry, where artificial intelligence, IoT and automation systems are effectively applied and are completely changing management. From water quality management to monitoring the health of farm animals, harvesting, and feeding are all done automatically and quickly with high efficiency. Among them, fish classification is an essential task in smart aquaculture. In the field of image classification DCNN has achieved remarkable successes. However, applying DCNN models also requires a large amount of data to train the model and a long training time. Therefore, applying a model based on MobileNetv2 has solved these problems. In this article, the fish image classification method based on MobileNetv2 with Transfer learning is applied. We used the MobileNetV2 network pre-trained by the ImageNet dataset as the base network and added the following layers to the base model and Softmax classifier, this new model is Adam optimized. This method finally achieved a classification accuracy of 98.18% in the test dataset of 330 images of 5 fish types.