Published Paper Details:

Agricultural productivity is a key component of the economy, but every year crops succumb to several diseases. Artificial Intelligence (AI) models for plant disease detection frequently struggle in real-world farm environments, primarily due to environmental noise and complex backgrounds that fail to capture the pristine variability of lab conditions. This study proposes an innovative, robust framework that leverages the compound scaling of EfficientNetB0 trained on raw, real-environment images to address this domain gap. The performance of EfficientNetB0 is systematically compared against DenseNet121 and MobileNetV2. To further bridge the gap between automated detection and practical agronomy, this system integrates the Gemini API to provide Explainable AI (XAI) and dynamic treatment remedies. The core objective is to develop a reliable, generalizable system that overcomes the "black-box" limitations of traditional CNNs, enhancing the accuracy (achieving 94% in field conditions) and efficiency of plant disease diagnosis in the unpredictable conditions of a real farm.