Digital Image-Based Chili Quality Detection Using a Web-Based Convolutional Neural Network

Abstract

ABSTRACT Chili is one of the main horticultural commodities in Indonesia, with high economic value and stable market demand. Accurate determination of chili quality levels is an important factor in maintaining quality, selling price, and distribution efficiency. Until now, the process of assessing chili quality has generally been carried out manually through direct visual inspection by experts or field officers. This traditional approach has limitations, such as varying levels of accuracy due to assessor subjectivity and the limited availability of experts. Advancements in digital image processing technology, particularly deep learning, offer opportunities to develop more accurate and consistent automated detection systems. This study proposes a Convolutional Neural Network (CNN) model to classify chili quality levels based on digital images, which is then integrated into a web-based application. This study uses a dataset of 405 chili images from 11 varietal categories, each labeled with quality (good, pest-infested, or unknown), which undergoes preprocessing stages including resizing, normalization, and data augmentation. The CNN model was designed with convolutional layers, max-pooling, dense layers, and a Softmax activation function, and was trained using the Adam optimizer and Categorical Cross-Entropy Loss. The web application implementation was carried out using the Flask framework, allowing users to upload images and obtain prediction results in real time. The testing results showed that the developed CNN model achieved an accuracy of 1.000 on the test data, with reliable detection performance under variations in lighting and image backgrounds. This research contributes to the development of smart agriculture technology by providing an accurate, fast, and easily accessible solution for chili quality detection