Intelligent System for Crop Disease Identification and Solution Recommendation Using CNN and EfficientNetB0

Abstract

The global food supply chain depends heavily on agricultural productivity. Crop diseases, however, still result in significant yield losses, particularly in rural areas where farmers have little access to professional diagnosis. An intelligent web- based system that uses deep learning models to identify crop diseases from leaf photos and offer trustworthy treatment recommendations is presented in this study. A publicly accessible dataset (Samir Bhattarai, Kaggle) with roughly 54,000 photos of different crop types and disease categories was used to train and compare two architectures: Convolutional Neural Network (CNN) and EfficientNetB0. While the Efficient NetB0 showed better performance with higher 95% training accuracy and 96% validation accuracy and better generalization, the CNN achieved 98% training accuracy and 95% validation accuracy. Multilingual audio alerts are integrated into the system (Marathi and Hindi) and web accessibility, making it inclusive and practical for rural farmers. Results indicate that EfficientNetB0 is more reliable and efficient due to its depth-wise scaling and parameter optimization, making it a suitable choice for real-time agricultural advisory systems.

KEYWORDS: Crop Disease Detection, CNN, EfficientNetB0, Deep Learning, Transfer Learning, Smart Agriculture, Web Application