Green AI Solutions for Automated Insect Monitoring to Improve Stored Product Management Systems

This study presents an integrated AI framework for automated beetle detection in stored product environments, leveraging EfficientNet Lite for feature extraction and XGBoost for classification. The system addresses the challenge of precise species identification in low-resource settings, enhancing postharvest sustainability. Our approach focuses on five stored product beetles: maize weevil (Sitophilus zeamais), red flour beetle (Tribolium castaneum), sawtoothed grain beetle (Oryzaephilus surinamensis), lesser grain borer (Rhyzopertha dominica), and rusty grain beetle (Cryptolestes ferrugineus). Experimental results demonstrate a training accuracy of 98.9% and a testing accuracy of 98.7<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup>. Additionally, a performance comparison with other models demonstrates superior resource efficiency. The hybrid attention map (HAM) enhances interpretability by localizing morphological features like antennae, elytra, and body contour. The proposed system achieved a performance optimization score (POS) of 239.2, indicating suitability for real-time IoT-enabled monitoring solutions. Our results indicate significant potential to reduce grain losses and minimize pesticide usage using this detection system.