Fish diseases detection using convolutional neural network (CNN)

Abstract

The fishing industry has become an important income source in the world. However, fish diseases are considered a serious problem among the fishermen as it tends to spread quickly through the water. In decades, fish diseases have been diagnosed manually by the naked eyes of experienced fish farmers. Despite being time-consuming since some lab works are required in determining the relevant microorganisms that cause the diseases, this classical method most often leads to an inaccurate and misleading result. Accordingly, a fast and inexpensive method is therefore important and desirable. Convolutional Neural Network (CNN) performance has recently been demonstrated in a variety of computer vision and machine learning problems. Thus, a study on fish diseases detection using CNN is proposed. A total of 90 images of healthy leaf and two types of fish diseases which are White spot and Red spot was tested. The application of CNN to a variety of testing datasets returned good detection accuracy at 94.44%. It can be inferred that the CNN is relatively good in detecting and classifying the type of diseases among infected fishes. Regardless, a study with a better number of datasets could be done in the future to improve the detection performance.