Implementation of Hybrid Machine Learning Techniques for Detection and Classification of Leaf and Stem Pests in Rice Crops

Abstract

Rice is a staple crop crucial to food security, particularly in Southeast Asia, where pest infestations cause substantial yield losses. In the Philippines, rice fields are highly susceptible to leaf and stem pests, which compromise productivity and farmers’ livelihood. Traditional pest monitoring methods are labor-intensive and error-prone. Although models like Pest-Net have reached 88.6% accuracy, limitations remain in real-time detection accuracy. This study presented a hybrid deep learning model integrating Convolutional Neural Networks (CNN) for feature extraction and YOLOv5 for real-time object detection and classification. A dataset containing eight rice pest species underwent augmentation and was evaluated using standard detection metrics. The proposed model achieved a mAP50 of 96.8%, significantly outperforming Pest-Net. Integrated into a GUI, the system enables real-time detection with class labels and confidence scores. This solution enhances precision agriculture in pest monitoring. Future work includes expanding pest class coverage and optimizing the system for deployment in diverse environmental settings.