<![CDATA[The manual grading of chili pepper dryness is uneven because human graders tend to perceive color, texture, and form changes that occur gradually during drying in a subjective way. The purpose of this study is to develop a lightweight convolutional neural network model that can effectively balance classification accuracy, validation stability, and deployment feasibility for inexpensive post-harvest sorting. A controlled visual dataset of 1,662 photos of red chili pepper from 32 samples at eleven drying times was gathered and classified into Fresh, Medium, and Dry classes. We assessed MobileNetV2, NASNetMobile, and InceptionV3 using the same pre-processing, augmentation, and hold-out testing protocol, along with additional robustness analysis. MobileNetV2 achieved the best hold-out performance with 93% accuracy, 93% precision, 92% recall, and 92% F1-score, while having fewer parameters and lower computational cost than NASNetMobile and InceptionV3. The class-wise analysis showed that the greatest errors were found between the Fresh–Medium and Medium–Dry boundaries, as the visual transition of chili dryness is gradual. MobileNetV2 is the most suitable baseline for low-cost camera-based chili pepper dryness sorting, and this study provides an evidence-based standard for post-harvest visual inspection using compact deep learning.]]>
Copyrights © 2026
