Classification of Different Types of Koi Fish Using Convolutional Neural Network

This study focuses on classifying different types of Koi fish using Convolutional Neural Network (CNN). The previous study conducted by Sapphira et al. (2020) used Zero Parameters Simple Linear Iterative Clustering (SLICO) for the object segmentation process and SVM as a classifier on identifying 15 types of Koi fish. Wherein, training and testing datasets are composed of 1200 and 300 images. These images of Koi fish are acquired from the Internet. Also, in this study, the proponents tested three colorspaces on getting the textures needed from the image and these are RGB, HSV, and grayscale colorspaces. In the segmentation process of their system, some images that have light reflections, shadows, water reflection, or water ripples become a major factor and cause the images to be unable to segment optimally. In relation to this, the average accuracies for using grayscale, RGB, and HSV are 36%, 50%, and 48%. In order to improve the accuracy in classifying the types of Koi fish, CNN, a deep learning algorithm is used. CNN has layers that are trained in such a way that it detects patterns and further along with more complex patterns. A convolution is inspecting the whole image for features needed to achieve higher accuracy of the predicted outcome. By using CNN, an average accuracy of 84% is computed from the testing conducted to the system.