LEAF-LEVEL CROP HEALTH ASSESSMENT USING DEEP LEARNING AND MOBILE IMAGE CAPTURE

Abstract

Timely detection of unhealthy crops is crucial to prevent yield reduction and to pro- mote precision agriculture. This study introduces a deep learning-based leaf health evaluation system employing the EfficientNetB0 architecture to classify crop leaves into two groups: healthy and diseased. RGB images similar to those captured with mobile devices serve as input, making the system suitable for real-time agricultural applications. Transfer learning, image enhancement, and incremental fine-tuning were implemented to improve model performance. Experimental findings indicate that the proposed system achieves an accuracy of approximately 95%, surpassing previous CNN-based methods. The approach is efficient, cost-effective, and can be deployed on mobile or edge devices, enabling farmers to perform quick self-assessments without expert intervention. This system supports smart farming practices by facilitating timely preventive measures and improving crop sustainability.