This research project aims to fill the gaps in contemporary smart agriculture (SmartAg) technologies by proposing a fog-assisted framework, FogAg, that integrates multi-layer sensing and real-time analytics of a plant-soil system to help solve the complex biological puzzle of linking the effect and interaction of two important crop inputs, viz., water and nitrogen, affecting crop yield. FogAg will also help provide near-real time diagnosis of crop stresses and translate the data into usable agronomic decisions not only to boost crop productivity but also to increase overall yield per unit of resource used in farming systems.
To meet the project goals and develop the proposed FogAg framework, scientific innovations in core CPS areas will be made on architecture, sensing, data analytics and machine learning, and modeling fronts. On the system-level CPS architecture front, the proposed FogAg architecture enables near real-time handling of latency-sensitive events in agriculture. On the device-level architecture, the project proposes Neuro-Sense, which will greatly support real-time signal/image processing by reconfiguration of data acquisition, computation, and communication parameters to assure energy-efficient performance for dynamically changing workloads. On the sensing front, the project proposes a sensing and imaging system for in-soil, above, below, and within plant canopy sensing. The proposed light emitting diode (LED) based multispectral imaging system will not only be economical but will also provide considerable flexibility by allowing different wavelengths to be used or substituted depending on measurement requirements as opposed to using filter-based methods common in multispectral systems or very expensive hyperspectral cameras. The proposed near-infrared (NIR) point measurement sensor will not only be an order of magnitude cheaper than a conventional diode array spectrometer but will also provide several options for interfacing and customizable optics. For in-soil sensing, a novel frequency response (FR)-based dielectric sensor and a mobile wireless sensor network are proposed that can simultaneously provide more accurate and real-time readings of soil nutrients and water content as compared to other commercially available soil sensors. On the data analytics and machine learning front, the proposed Neuro-Sense integrates a convolutional neural network (CNN) accelerator that will provide a higher throughput, utilization, area, and energy efficiency as compared to existing CNN accelerators by exploiting two-sided sparsity, binary sparse masks, and capability to look ahead into future computations to determine only the valid multiply and accumulate operations. The proposed three-tier data analytics FogAg framework will enable processing of sensing data at three levels: IoT, fog, and cloud. On the modeling front, tree-based predictive agronomic models will be developed that will utilize multi-layer sensed data along with historical yield and weather data to provide a variable-rate fertilizer and irrigation prescription map in near real-time to boost productivity and yield per unit of resource used in the farming system.
The proposed FogAg framework with spatial and temporal scalability will find many applications in rural and urban development. The proposed technologies will help in efficient usage of resources and improvement in crop health, quality, and yield thus creating significant social and economic benefits in food security. The joint consideration of water-nitrogen colimitations will have positive environmental impacts by reducing the nitrogen footprint and environmental pollution. The investigators will incorporate significant research results obtained during this project into undergraduate and graduate education. For participation of underrepresented multicultural students in the project, the PIs will collaborate with KSU Multicultural Engineering Program (MEP) for supporting broadening participation in computing (BPC) and engineering (BPE) activities.