International Journal of Advanced Research in Computer and Communication Engineering

Abstract: Deep learning has become a transformative approach for automated plant health monitoring, enabling accurate disease recognition directly from leaf images without relying on manual inspection or expert availability. In modern precision agriculture, both leaf disease classification and severity quantification are essential for identifying early infections and supporting informed intervention strategies. However, developing reliable diagnostic models remains challenging due to environmental variability, heterogeneous field conditions, inconsistent image quality, and the absence of pixel-level severity annotations in standard datasets. This literature-aligned study synthesizes advances in lightweight CNN architectures, classical image-processing pipelines, attention-guided visualization tools, and mobile-centric deployment frameworks for real-time plant disease assessment. Special emphasis is placed on the proposed end-to-end system, which integrates a custom PyTorch-based CNN with an OpenCV-driven severity estimation module and a cross-platform React Native mobile interface. While originally optimized for binary classification, the system directly addresses practical agricultural constraints such as uneven lighting, morphological variations across species, and limited computational resources in field environments. By combining interpretable predictions, severity mapping, and rapid inference via a Flask backend, the approach enhances usability, improves generalization under diverse conditions, and reduces diagnostic dependency on experts. Through comparative analysis with existing methods, this work positions the proposed framework as a promising foundation for future mobile plant-disease diagnostic pipelines integrating accessibility, explainability, and deployment-scale robustness.

Keywords: Plant Disease Detection, Convolutional Neural Networks, Severity Estimation, Mobile Application, Deep Learning, Precision Agriculture.

Downloads: | DOI: 10.17148/IJARCCE.2025.141244