LIGHTWEIGHT PLANT DISEASE DETECTION FRAMEWORK USING MOBILENETV2 AND TRANSFER LEARNING

Agriculture is the backbone of economies across the world, particularly in developing nations where a significant portion of the population depends on crop cultivation for livelihood. Plant diseases represent one of the most serious threats to agricultural productivity, causing between 20% and 40% of global crop yield loss annually. Timely and accurate identification of plant diseases is critical to enabling farmers to take corrective measures before losses escalate. However, traditional disease diagnosis relies heavily on expert agronomists, which creates bottlenecks due to limited specialist availability, especially in rural and remote farming regions. Plant Disease Prediction using Deep Learning is designed as a unified, intelligent agricultural assistance platform that enables automated identification of 38 distinct plant disease categories from leaf photographs with high accuracy. The system uses MobileNetV2 transfer learning, trained on the PlantVillage dataset containing 15,000 labeled leaf images spanning 14 plant species and 38 disease classes. This multiclass goal is important because real-world agricultural environments involve a wide variety of crops, and a scalable AI platform must adapt across different plant species and disease types. The project addresses a fundamental machine learning limitation in the agricultural domain: the challenge of achieving high accuracy with limited labeled data through efficient model architectures. By leveraging MobileNetV2, a model pre-trained on ImageNet's 1.2 million images, the system benefits from powerful feature representations without training from scratch. In project phase II, the work extends beyond model research to focus on productization: a Flask REST API backend, an interactive web application, Grad-CAM visual explainability, and real-time inference with sub-10ms response times. 1.1 BACKGROUND The result is not just a model demonstration but a complete system architecture. The platform includes a webbased frontend, a Flask REST API backend, and an integrated Grad-CAM explainability module. This makes the project suitable for academic demonstration, software engineering evaluation, and future deployment as a clinically assistive precision agriculture tool.