Analysis of CNN Method for Image Classification of Coconut Ripeness Levels

This research aims to overcome problems in the production of Pliek U Aceh, namely the difficulty in choosing the right coconut to create an optimal mixture, especially in distinguishing mature coconut, half-ripe coconut, and old coconut. To achieve this goal, Convolutional Neural Network (CNN) methods are explored using three different architectures, namely ResNet50, VGG16, and the underlying CNN architecture. The CNN model is used for coconut image acquisition, image preprocessing, CNN model training, and model testing on coconut image data. The problem studied was the inability of practitioners to select coconuts with high enough accuracy to produce an optimal mixture in Pliek U Aceh. Half-ripe coconuts are often difficult to distinguish from mature or mature coconuts, especially after being exposed to sunlight. Therefore, a model is needed that can identify coconuts with high accuracy to improve the quality of Pliek U. The proposed method includes coconut image acquisition, followed by an image preprocessing process to improve the quality of the input data. Then, CNN models were trained using the prepared dataset to classify coconuts into three categories: ripe, semi-ripe, and old. The evaluation was carried out to compare the performance of three different CNN architectures, namely ResNet50, VGG16, and the basic CNN architecture. The research results show that using the CNN model achieves an accuracy of 79%, while VGG16 has an accuracy of 41%, and ResNet50 has an accuracy of 56%. These results indicate that the use of the CNN method has great potential to increase accuracy in classifying coconuts according to the categories required for the production of Pliek U Aceh