A Comparative Study of Deep Learning-Based Pest Detection Models with Interpretability and Deployment Strategies for Medicinal Plants
Pest infestations significantly affect the growth, quality, and commercial value of medicinal plants, and early detection is essential for effective crop management. Manual inspection of pest infestation is labor-intensive, subjective, and error-prone. Although deep learning has shown significant performance improvements in plant disease detection, studies on medicinal plants, with an emphasis on interpretability and deployment feasibility, remain limited. This paper presents a comparative analysis of five deep learning models, namely Basic CNN, VGG16, ResNet50, VGG32, and the proposed CFNET, which is based on EfficientNetB3, for binary pest detection in medicinal plants. Experiments were conducted on curated datasets of healthy and infected leaf images. The models were evaluated using accuracy, precision, recall, F1-score, and training time. Among all models, the proposed CFNET achieved a peak accuracy of 0.98 and consistently outperformed the baselines across all datasets. To enhance transparency, explainable AI techniques, including Grad-CAM, Grad-CAM++, LIME, and SHAP, were employed. CFNET produces focused, visually interpretable explanations consistent with known infection patterns. Deployment analysis further confirms CFNET's suitability for mobile and edge-based agricultural monitoring systems.