Leveraging AI to Study the Impact of Climate Change on the Spread and Severity of Plant Diseases

Abstract

Rapid and accurate disease detection is essential for effective plant health management, which significantly increases crop yields and mitigates losses exacerbated by climate change. While machine-based imaging can detect symptoms invisible to the naked eye, standard models often overlook the influence of climatic and soil factors on disease occurrence. 

Objectives: To develop and assess the performance of deep learning models employing multimodal fusion strategies for the precise classification of crop diseases. 

Methodology: Proposed multimodal deep learning approach to combine visual features using MobileNetV2 with environmental and soil, including temperature, humidity, pH, rainfall, and nutrient levels (N, P, K) by the Multilayer Perceptron (MLP). 

Results: Experimental results demonstrate that MobileNetV2 excels among image-only models, proving to be an efficient lightweight architecture for agricultural tasks. Notably, the multimodal MobileNetV2 + MLP model achieved nearly 100% precision, significantly outperforming unimodal models. This highlights that incorporating environmental variables substantially enhances classification accuracy. 

Conclusion: Disease detection may become more precise if the multimodal correlations are strengthened by extending the dataset and upgrading the quality of environmental features, advanced fusion and alignment techniques usage that can further facilitate the interaction between the image and structured data, and lightweight CNNs, along with real-world testing that can help in the creation of reliable field-ready plant disease detection systems.