Deep learning and GIS integration for plant disease diagnosis and management

Abstract Plant diseases pose a major threat to global food security, particularly in agricultural regions such as Egypt. This study presents an integrated framework that combines deep learning (DL) with Geographic Information Systems (GIS) for the automated diagnosis and spatial monitoring of tomato and potato diseases. Three deep learning models were developed and evaluated: DenseNet169, MobileNetV2, and a Custom-CNN baseline. Comparative experiments on the test dataset showed that DenseNet169 achieved the highest evaluation accuracy of 98.24%, followed by MobileNetV2 with 94.66%, while the Custom-CNN obtained a lower accuracy of 82.60%, reflecting the limitations of shallow architectures trained from scratch. Owing to its superior performance, DenseNet169 was selected for integration with the GIS module. The GIS component geotags disease predictions and visualizes outbreaks through interactive crop-specific layers, heatmaps, and temporal tracking, enabling early detection and targeted interventions. This spatial–temporal integration allows stakeholders to monitor disease progression, assess risk under varying environmental conditions, and optimize resource allocation for precision agriculture. The final system also incorporates a bilingual web interface (Arabic/English) enhanced with Grad-CAM visualizations, which improve interpretability and build trust among local farming communities. Field validation conducted in Beheira Governorate, Egypt, confirmed the accuracy and practical utility of the system in real-world agricultural settings. By integrating AI-based diagnostics with geospatial intelligence, the proposed framework provides a scalable and sustainable decision-support tool, offering a valuable pathway toward data-driven plant disease management in developing regions.