Sensor Fusion And Machine Learning Algorithms for the Proactive Classification of Early Stage Agricultural Diseases

Abstract

In this paper, Hybrid-PAIA, the new system of presymptomatic detection of crop diseases based on intelligent multimodal sensor fusion, is introduced. The framework combines multispectral imagery collected by the UAV with in-situ IoT environmental information by dynamically setting priorities on the diagnostic features of each modality with a cross-modal attention mechanism. On tomato and wheat data, Hybrid-PAIA was found to have classification accuracy of 96.4% and early detection at 88% and detected infections 4.2 days prior to observable symptoms. Significant correlation between weeds and disease hotspots was found in the system (r=0.67), as well as the error of prediction of yield was minimized (51.9). It is computationally efficient as it can process one hectare in 4.8 seconds using edge hardware and is evidence that intelligent sensor fusion can make disease management an active, precision instrument to the sustainable agriculture process.