Improving Classification and Clustering Accuracy of Palembang Duku Fruit Quality via CNN-K-Means Integration

Abstract

Duku Palembang (Lansium domesticum Corr.) is a leading commodity from South Sumatra with high economic value. The quality assessment process for duku has traditionally relied on manual visual inspection, which is subjective, time-consuming, and inconsistent, especially at large production scales. This study proposes an integrative framework combining Convolutional Neural Network (CNN) and K-Means Clustering to improve the accuracy of classification and clustering of Duku Palembang quality. CNN is employed to extract high-level visual features such as color, texture, shape, and edges, while K-Means utilizes feature similarity to group fruits in an unsupervised manner, maintaining effectiveness even when labeled data is limited. The research stages include image data preprocessing, feature extraction using CNN, supervised quality classification, and quality grouping using K-Means. Experimental results show that integrating the two methods enhances the accuracy of duku quality identification compared to using a single method, while also offering a fast, objective solution that can be implemented in sorting and distribution facilities. This approach is expected to support precision agriculture practices and increase the competitiveness of Indonesian agricultural products in the global market.