Classification of Rastrelliger kanagurta and Rastrelliger brachysoma using Convulational Neural Network (CNN)

Abstract

Mackerel is an important commercial caught fish for local fishermen, including Rastrelliger kanagurta and Rastrelliger brachysoma. However, to distinguish these two species is rather difficult because of their similar appearance. Convolutional Neural Network (CNN) is a deep learning method that can be used to classify images. One of parameters contributing to the level of accuracy is the layers number applied in CNN architecture. This study aims to classify those two species using CNN with a range of two to five convolutional layers architectures i.e CNN1, CNN2, CNN3 and CNN4, respectively. In this study, 434 images were used as a training group with 217 images for each class. The validation group consisted of 21 images for each class and the test group consisted of 19 images. The results showed that the CNN3 provided the best training and validation accuracy of respectively 100% and 92.6%. The lowest value of training loss and validation loss of 0.000057 and 0.49. The accuracy values of the CNN models using different testing images reached 94.7%.