Forecasting Sterility Mosaic Disease in Pigeonpea Using Dynamic Bayesian Networks and 3D Point Cloud High-throughput Scanning Platform

This paper explores how high-throughput phenotyping can be integrated with machine learning models to efficiently forecast the Sterility Mosaic Disease using a small amount of training data. This approach is generalized through the use of a Dynamic Bayesian Network (DBN). To predict the spread of the virus, the entire network is decomposed into several distributed and cooperative learning modules. The EM algorithm is used to learn the parameters for each module. Upon iterative convergence, the estimated hidden state vector of one module serves as input control for the next. The parameter estimates of the final module are used to formulate a predictor capable of forecasting q-days ahead.To demonstrate the effectiveness of the proposed DBN, its performance is evaluated using real-world data from ICRISAT, Patancheru, Hyderabad, Telangana, India. Physiological data was collected using 3D point cloud technology, while environmental data was recorded by a local weather station.