Determining Tuna Grade Quality Based on Color Using Convolutional Neural Network and k-Nearest Neighbors

Abstract

One of the main commodities that Indonesia exports is tuna. Indonesia's inadequate handling of food safety is demonstrated by a number of instances when the United States has rejected Indonesian fishery goods and food poisoning incidences. Fish quality grade is currently determined by manual inspection which has risk human mistake. According to Robert DiGregorio, four tuna grade classifications exist: grade 1, 2+, 2, and 3. The purpose of this study is to assess the tuna meat's quality according to its color. The procedure involves pre-processing images, training datasets, and classifying them using the Convolutional Neural Network (CNN) and k-Nearest Neighbors algorithms. CNN pre-processing involves converting the image into HSV color space and training the CNN model using 240 training datasets and 74 testing datasets. CNN’s accuracy was 84% higher than k-Nearest Neighbors' which was 54%. Additionally, a comparison of the classification accuracy of CNN, VGG (Visual Geometry Group) 16, and AlexNet revealed that CNN outperformed the others with an accuracy of 84%, followed by VGG16 with 70% and AlexNet with 66%.