Intelligent Multicrop Disease Detection Using Adaptive Multimodal Hybrid Deep Learning: A Comprehensive Framework For Indian Agricultural Systems

Abstract

Crop diseases pose a significant threat to agricultural productivity in India, with traditional detection methods being time-intensive, subjective, and often inaccurate. The diversity of crops and disease patterns across different agro-climatic zones necessitates an intelligent, adaptive approach to disease detection. This study proposes and validates an innovative multimodal hybrid deep learning framework that automatically selects optimal model architectures for detecting diseases across multiple Indian crop varieties using drone-acquired multispectral imagery. We developed an adaptive ensemble framework combining Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and EfficientNet architectures with a novel Meta-Learning Model Selector (MLMS) that dynamically chooses the best-performing model combination for specific crop-disease scenarios. The system was validated across 1,247 agricultural plots covering five major Indian crops, processing 67,500 high-resolution multispectral images collected via drone surveys across Maharashtra, Punjab, and Tamil Nadu during 2022-2024. The proposed Adaptive Multimodal Hybrid Network (AMHN) achieved superior performance with 96.3% overall accuracy (95% CI: 94.8-97.6%), significantly outperforming individual architectures. The Meta-Learning Model Selector demonstrated 98.7% accuracy in selecting optimal models for specific scenarios. The framework successfully detected six major disease categories with precision values ranging from 94.2% to 97.8% across different crops. The validated multimodal hybrid framework provides a robust, scalable solution for automated crop disease detection, offering significant improvements in accuracy and adaptability compared to single-model approaches.