<![CDATA[Rice is a key commodity in national food security, yet its productivity often declines due to pest attacks. Manual pest identification is considered inefficient, necessitating an accurate automated system. This study proposes an image classification model for rice crop pests using a Convolutional Neural Network with the MobileNetV3-Small architecture through a transfer learning approach. The model development process follows the SEMMA methodology (Sample, Explore, Modify, Model, Assess). The dataset consists of 5,395 images across seven major pest classes—brown planthopper, green planthopper, rice stem borer, larvae, thrips, false white pest, and rice water weevil—sourced from Kaggle, along with additional test data obtained through web scraping to simulate real-world conditions.Image quality analysis revealed variations in blur levels and lighting across classes, which were addressed using augmentation and class weighting. The model was trained with fine-tuning on upper layers, using the Adam optimizer and early stopping over 30 epochs. The results show a training accuracy of 94.18%, validation accuracy of 97.58%, and test accuracy of 98%, with average precision, recall, and F1-score values of 0.98. A 5-fold cross-validation yielded an average accuracy of 97.13% with a deviation of ±0.81, indicating stable performance. Compared with MobileNetV2, the MobileNetV3-Small model performed significantly better in both accuracy and computational efficiency (p-value = 0.0269). These findings demonstrate that lightweight architectures such as MobileNetV3-Small are effective for rice pest classification and hold potential for implementation in automated detection systems for smart agriculture applications.]]>
Copyrights © 2026
