Vol. 7, Issue 9, Part A (2025)

The growing occurrences of multi-stage crop pathogens due to climatic vagaries and agricultural encroachment have necessitated the implementation of early warning systems of countering the crop diseases. In this work, a Deep Transfer Learning-based, multi-modal sensor input predictive modelling framework incorporating deep transfer learning and multi-modal sensor input data assembly is developed to determine crop diseases at different phase of growth. The suggested method is a combination of hyperspectral images and soil nutrient sensors and weather stations to construct a powerful temporal-spatial model. Transferred learning is being used to train pre trained convolutional neural networks with task specific agricultural data which increases the efficiency of the model and it shortens the training time. It has temporal attention mechanisms and modelling of disease progression to identify the slight shifts in disease conditions. The system is better in accuracy, allowing precision farming insight action. In addition, the model can be deployed in a scalable manner throughout the edge-AI platforms in real-time monitoring and control. They are more accurate than other approaches as the suggested system had the highest rates of accuracy at 96.8% to detect crop disease at multiple stages., and thus it is a very efficient tool to implement in the active control of crop diseases. The work will be part of the future of sustainable agriculture since it will reduce the loss of output and optimize the utilization of resources by enabling early, precise, and automatic identification of the disease.