A Multi-Scale Data Assimilation Framework for Layered Sensing and Hierarchical Control of Disease Spread in Field

This project is focused on developing data analytics and decision-making techniques for early detection and mitigation of soybean diseases via fusing data from ground robots, UAVs and satellites. We aim to collect RGB and hyperspectral image data for soybean diseases from research farms at Iowa State and via collaboration with the Iowa Soybean Association and the NASA Jet Propulsion Lab (for satellite data). Upon data collection, we will develop a machine learning framework to efficiently fuse multi-resolution and multi-spectral information from ground robots, UAVs and satellites for early detection of a critical soybean disease and estimate the disease severity progression as well as spatial progression of the disease. Such a framework for disease identification, severity quantification and prediction will enable us to develop mitigation strategies that farmers can use to reduce the impact of the disease.

In this project, we have developed a deep learning-based vision framework that is capable of detecting and quantifying the soybean diseases by looking at visual symptoms on soybean leaf images. This framework can be deployed as an app on a smartphone or tablet - so that farmers are able to detect onset of diseases early and take appropriate steps. We have also developed a 3D CNN framework for early detection of a soybean disease using hyperspectral data. Towards crop-loss mitigation, we also put forward a Deep Learning-based novel approach for accurately detecting, counting and estimating yield of sorghum crops from crop row images. We aim to design a detection model capable of efficiently detecting, counting crop heads and give an accurate estimate of yield. We also, further use the developed model for auto-annotating new data across different genotypes irrespective of shape and/or size of the sorghum heads.

This project is supported by the USDA-NIFA under Grant No. 2017-67007-26151.