A MULTI-SCALE DATA ASSIMILATION FRAMEWORK FOR LAYERED SENSING AND HIERARCHICAL CONTROL OF DISEASE SPREAD IN FIELD CROPS

• Sponsoring Institution: National Institute of Food and Agriculture
• Project Status: COMPLETE
• Funding Source: AFRI COMPETITIVE GRANT
• Reporting Frequency: Annual
• Accession No.: 1011790
• Grant No.: 2017-67007-26151
• Cumulative Award Amt.: $990,471.00
• Proposal No.: 2016-11395
• Project Start Date: Mar 1, 2017
• Project End Date: Feb 28, 2021
• Grant Year: 2017
• Program Code: [A7302]- Cyber-Physical Systems

Recipient Organization
IOWA STATE UNIVERSITY
2229 Lincoln Way
AMES,IA 50011

Performing Department
Mechanical Engineering

Non Technical Summary
The availability of cheap, deployable, and connected sensor technology has produced a data deluge at varying spatial, temporal, and fidelity scales. Currently, algorithms are not well-suited to assimilate multi-scale heterogeneous data, such as continuous temporal data with discrete event driven data, or image data with time-series data; therefore a major challenge is the lack of standardization which would facilitate the interoperability of different platforms.The objective of this proposal is to develop and deploy an agricultural analytics framework that assimilates multi-scale heterogeneous data and provides actionable information for the control of disease spread in field crops. The proposed framework will (a) effectively extract actionable information from multi-length/time/frequency scale data at the plant, plot and field levels; and (b) build an efficient decision support system for early detection and mitigation of row crop diseases. We will accomplish the objectives via: (a) using machine learning concepts to extract disease signatures from heterogeneous data within a layered sensing architecture; (b) developing graphical, and probabilistic models for multi-scale data assimilation and prediction of localized disease onset and spread; (c) integration of mathematical, and cyberinfrastructure tools with physical multi-scale sensing platforms for hierarchical active information gathering and subsequent decision-support for mitigation strategies; and (d) deployment of the framework on research farms as well as dissemination of results to the broader farming and research community.The outcomes of this research will benefit a broad spectrum of the agricultural community, from small scale farmers to large scale farming operations and plant scientists. The validated framework will facilitate informed decision-making by the farming community and to accelerate plant breeding to create robust crops under changing climatic conditions. We will package these developments into modular, web-based, and 'app' based tools that are deployable in freely available, cloud-based cyberinfrastructures. The highly interdisclipinary research activities will help train the next generation of scientists and engineers to tightly integrate cyberinfrastructure advances with agricultural systems.

Animal Health Component

40%

Research Effort Categories

Basic

50%

Applied

40%

Developmental

10%

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
212	1820	2080	20%
212	1820	2020	20%
205	1820	2020	20%
203	1820	1081	10%
203	1820	1160	15%
212	1820	1160	15%

Knowledge Area
212 - Pathogens and Nematodes Affecting Plants; 205 - Plant Management Systems; 203 - Plant Biological Efficiency and Abiotic Stresses Affecting Plants;

Subject Of Investigation
1820 - Soybean;

Field Of Science
2020 - Engineering; 1160 - Pathology; 1081 - Breeding; 2080 - Mathematics and computer sciences;

Keywords

cyber-agriculture system

decision support

disease detection and mitigation

Goals / Objectives
The proposed project has two broad goals:Goal 1: Bottom-up hyperspectral data assimilation, compression, and signature identification to integrate ground based hyperspectral images, multispectral data collected by UAVs, and hyperspectral satellite datato identify unique spectral signatures of various soybean diseases. The specific objectives related to this goal are:Objective 1 - Building a layered sensing architecture (FTE requirement - 0.6)Objective 2 - Create disease signature analytics using multi-modal data (FTE requirement - 1.2)Objective 3 - Development of a neural-symbolic framework for spatiotemporal data modeling (FTE requirement - 0.7)Objective 4 - Causal graphical modeling of disease spread dynamics (FTE requirement - 1.2)Goal 2: Top-down hierarchical control and deployment to localize plant stress signatures and estimate disease severity along with providing actionable information enabling early stage disease mitigationObjective 5 - Hierarchical active information gathering (FTE requirement - 0.7)Objective 6 - Real-time planning of mitigation strategies (FTE requirement - 1)Objective 7 - Demonstrate the efficacy of the multiscale decision support framework on field scale (FTE requirement - 2.5)Objective 8 - Buildinga large scale deployment framework (FTE requirement - 1)

Project Methods
To enable real-time assimilation of multi-scale, multi-modal data including hyperspectral images, we propose to build a novel neural symbolic framework. Recent advancements in deep learning shows that neural approaches are excellent at low-level feature extraction from raw data, automated learning and discriminative tasks. However, such models still are not well suited for logical reasoning, interpretation, and domain knowledge incorporation. Symbolic approaches can alleviate such issues as they are effective in high-level reasoning and capturing sequence of actions. Therefore, a hybrid neural-symbolic learning architecture has the potential to execute high-level reasoning tasks using the symbolic part based on the automated features extracted by the neural segment. Hierarchical feature extraction using deep neural networks very successful provided state-of-the-art results for various machine learning tasks. Analyzing very high volume hyperspectral data using 3D Convolution Neural Nets (CNN) in particular offers the potential to detect plant diseases prior to the detection of visible symptoms. After the extraction of neural features, we will use a symbolic framework to model disease progression and spread dynamics.Finally, themodeling framework will be used for real-time decision making to enable farmers to actively collect critical data for early detection of diseases on their farms. The system will also recommend mitigation decisions.Efforts: Please read the 'Services' section of the 'Products'.Evaluation: Performance evaluationof proposed cyber-agricultural system will be conducted using quantifiable metrics such as,1) Compression measure: the efficacy of this framework in terms memory/storage reduction resulting from the identification of spectral bands. 2) Reduction in time to detect/control efficacy: impact of the control action in terms of reduction in time to detect a specific set of diseases. 3) Sustainability gain: % reduction in spread/volume of chemical treatment required for effective mitigation of the disease. 4) Gain in efficiency: quantification of yield improvement - how many additional bushels/area does this framework produce.

Progress 03/01/17 to 02/28/21

Outputs
Target Audience:During the entire durationof this project, we communicated our research outcomes to the broad scientific research community via multiple publications in reputed journals and conferences. Apart from this, we have also presented our work in the NSF CPS PI meetings and in the US-Japan collaboration meetings in Tokyo. Apart from this, we have also presented our work in multiple invited talks at various international workshops,conferences, companies and universities. Our work has also reached the soybean farmers community (primarily in Iowa) and agriculture companies such as John Deere, Monsanto, Corteva through various meetings with the project director and co-directors. Changes/Problems:While we made significant technical progress to complete all the techncial goals of the project according to the proposal, detail experimental validation of the management and mitigation strategies and their deployment in agricultural research fields have suffered some roadblocks due to all the Covid-19 related disruptions. What opportunities for training and professional development has the project provided?The project has provided significant training and professional development opportunity to multiple graduate students and a postdoc researcher. The students (from both plant science and engineering disciplines) gained valuable interdisciplinary expertise to successfully conduct cyber-agricultural research. They also had professional development experience via presenting their work in conferences and workshops. How have the results been disseminated to communities of interest?The results have been disseminated through the publications mentioned in the "products" section of our four annual reports.Also, the students have delivered talks in various relevant venues such as the PAG,Phenome meetings,North American Plant Phenotyping Network (NAPPN) Annual Meeting and theNeurIPS Workshop on AI for Earth Sciences. The project directors have delivered a few invited talks, such as keynote addresses at the Workshop on Data Science for Agriculture and Natural Resource Management during the 8th International Conference on Big Data Analyticsand theJapanese Society of Agricultural Informatics (JSAI) Annual Conference.PD Sarkar delivered a Keynote address - Deep Learning for Agricultural Analytics, Keynote address in Asia-Pacific Federation for Information Technology in Agriculture/World Congress on Computers in Agriculture (AFITA/WCCA), IIT Bombay, Mumbai, India, October, 2018. Our team was also invited to the 2018 Farm Progress Show where we showcased our Soybean disease detection app work using multi-stresses detection app to midwestern farmers along with international farmers from brazil, Argentina and China etc. as well as to the to US Secretary of Ag Mr. Sonny Purdue. Our project "Artificial intelligence and crop scouting" is one of three picked projects to showcase Iowa State stories of statewide impact, innovation and entrepreneurship during the "2019 ISU Day" at the Iowa State Capitol on March 6, 2019. We also started the International Workshop Series on Machine Learning for Cyber-Agriculture Systems. We also co-organized a CPS-Ag workshop within the 2018 NSF CPS PI meeting. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? In the first year of the project, we focused on these following objectives - Objective 1 - Building a layered sensing architecture, Objective 2 - Create disease signature analytics using multi-modal data and Objective 5 - Hierarchical active information gathering. We have made significant progress in all three objectives. For Objective 1, we have set up hardware/software architecture and data gathering protocol for ground-based manual/robotic imaging and UAV based imaging. We have established a partnership with Planet Labs to gather relevant satellite data. For Objective 2, we have built and successfully demonstrated deep learning based multi-modal (RGB and Hyperspectral images) data analytics for plant disease detection. Finally, we have developed initial algorithms for optimally deciding the data necessary for row crop stress management. In the secondyear of the project, we focused on these following objectives - Objective 2 - Create disease signature analytics using multi-modal data, Objective 3 - Development of a neural-symbolic framework for spatiotemporal data modeling, Objective 4 - Causal graphical modeling of disease spread dynamics, Objective 7 - Demonstrate the efficacy of the multiscale decision support framework on field scale, and Objective 8 - Building a large scale deployment framework. We made significant progress on all these objectives. We built deep learning based models for analyzing both RGB and Hyperspectral information for plant disease identification (Obj 2). We also built an mobile app for Soybean disease detection using deep learning. We made algorithmic developments in the area of neural-symbolic modeling, Granger-causal graphical modeling and deep LSTM models for spatiotemporal data modeling (Obj 3,4). For large field scale deployment, we built a label-efficient deep learning framework that can be used efficiently for UAV based images (Obj 7,8). In the third year of the project, we focused on these following objectives - spatiotemporal data assimilation and predictive modeling for yield prediction, breeding support (Obj 3, 7), canopy-level stress identification in field using weakly supervised deep learning and active learning (Obj 3, 7) and robotic deployment of automated yield estimation (Obj 8). In the final year of the project, we focused on these following objectives - spatiotemporal data assimilation and interpretable predictive modeling for yield prediction, breeding support (Obj 3, 7), plant modeling using hybrid machine learning models that incorporates domian knowledge and crop modeling constraints (Obj 3, 6, 7), optimal management decision-making using reinforcement learning (Obj 6) and various advanced high throughput information gathering including complex plant phenotypic information (Obj 1, 5, 8).

Publications

Progress 03/01/20 to 02/28/21

Outputs
Target Audience:During the finalyear of this project, we communicated our research outcomes to the broad scientific research community via multiple publications in reputed journals and conferences. Apart from this, we have also presented our work in multiple invited talks at various international workshops and conferences in India and Japan.Our work has also reached the soybean farmers community (primarily in Iowa) and agriculture companies such as John Deere, Monsanto, Corteva through various meetings with the project director and co-directors. Changes/Problems:While we made significant technical progress to complete all the techncial goals ofthe project according to the proposal, detail experimental validation of the management and mitigation strategies and their deployment in agricultural research fields have suffered some roadblocks due to all the Covid-19 related disruptions. What opportunities for training and professional development has the project provided?The project has provided significant training and professional development opportunity to multiple graduate students and a postdoc researcher. The students (from both plant science and engineering disciplines) gained valuable interdisciplinary expertise to successfully conduct cyber-agricultural research. They also had professional development experience via presenting their work in conferences and workshops. How have the results been disseminated to communities of interest?The results have been disseminated through the publications mentioned in the "products" section earlier. Also, the students have delivered talks in various relevant venues such as the 2021 North American Plant Phenotyping Network (NAPPN) Annual Meeting and the 2020 NeurIPS Workshop on AI for Earth Sciences. The project directors have delivered a few invited talks, such as keynote addresses at theWorkshop on Data Science for Agriculture and Natural Resource Management during the 8th International Conference on Big Data Analytics 2020 and the 2021 Japanese Society of Agricultural Informatics (JSAI) Annual Conference. However, we do highlight the fact that the extent of dissemination was significantly reduced due to Covid-19 related disruptions. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? In the finalyear of the project, we focused on these following objectives - spatiotemporal data assimilation and interpretable predictive modeling for yield prediction, breeding support (Obj 3, 7), plant modeling using hybrid machine learning models that incorporates domian knowledge and crop modeling constraints(Obj 3, 6, 7), optimal management decision-making using reinforcement learning (Obj 6) and various advanced high throughput information gathering including complex plant phenotypic information(Obj 1, 5, 8).

Publications

• Type: Journal Articles Status: Published Year Published: 2020 Citation: Kevin G. Falk, Talukder Zaki Jubery, Jamie A. ORourke, Arti Singh, Soumik Sarkar, Baskar Ganapathysubramanian, Asheesh K. Singh, Soybean root system architecture traits study through genotypic, phenotypic and shape-based clusters, Plant Phenomics, 2020.
• Type: Journal Articles Status: Published Year Published: 2020 Citation: S. V. Mirnezami, T. Young, T. Assefa, S. Prichard, K.Nagasubramanian, K. Sandhu, S. Sarkar, S. Sundararajan, M. E. O'Neal, B. Ganaphathysubramanian, A. Singh, Automated trichome counting in soybean using advanced image?processing techniques, Applications in Plant Sciences, 2020.
• Type: Journal Articles Status: Published Year Published: 2020 Citation: A. Singh, S. Jones, B. Ganapathysubramanian, S. Sarkar, D. Mueller, K. Sandhu, K. Nagasubramanian, Challenges and Opportunities in Machine-Augmented Plan Stress Phenotyping, Trends in Plan Science - Cell Press, August 2020, DOI:https://doi.org/10.1016/j.tplants.2020.07.010?
• Type: Journal Articles Status: Published Year Published: 2020 Citation: Z. Jiang, C. Liu, B. Ganapathysubramanian, D. J. Hayes, S. Sarkar, Predicting county-scale maize yields with publicly available data, Scientific Reports, September 2020, DOI:https://doi.org/10.1038/s41598-020-71898-8
• Type: Journal Articles Status: Awaiting Publication Year Published: 2021 Citation: L.G. Riera, "Deep Multi-view Image Fusion for Soybean Yield Estimation in Breeding Applications", Plant Phenomics, 2021.
• Type: Journal Articles Status: Awaiting Publication Year Published: 2021 Citation: Nagasubramanian, Koushik et al. "How useful is Active Learning for Image-based Plant Phenotyping?" The Plant Phenome Journal (2021).
• Type: Journal Articles Status: Awaiting Publication Year Published: 2021 Citation: Wei Guo et al. "UAS based plant phenotyping for research and breeding applications" Plant Phenomics, 2021
• Type: Journal Articles Status: Awaiting Publication Year Published: 2021 Citation: Johnathon Shook, Tryambak Gangopadhyay, Linjiang Wu, Baskar Ganapathysubramanian, Soumik Sarkar, Asheesh K. Singh, "Crop Yield Prediction Integrating Genotype and Weather Variables Using Deep Learning", PLOS One, 2021
• Type: Conference Papers and Presentations Status: Published Year Published: 2020 Citation: T. Gangopadhyay, J. Shook, A. K. Singh, S. Sarkar, Interpreting the Impact of Weather on Crop Yield Using Attention, NeurIPS Workshop on AI for Earth Sciences, 2020.

Progress 03/01/19 to 02/29/20

Outputs
Target Audience:During the thirdyear of this project, we communicated our research outcomes to the broad scientific research community via multiple publications in reputed journals and conferences. Apart from this, we have also presented our work in the 2019NSF CPS PI meeting as a flash talk. Our work has also reached the soybean farmers community (primarily in Iowa) and agriculture companies such as Monsanto, Corteva through various meetings with the project director and co-directors. Changes/Problems:While we made significant technical progress in the project according to the proposal, detail experimental validation and deployment in agricultural research fields have suffered some roadblocks due to inclement weather and delay in acquiring some specialized sensors from vendors. Therefore, we are requesting a one year no-cost extension to complete the proposed experiments, validation and necessary software development and implementations. What opportunities for training and professional development has the project provided?The project has provided significant training and professional development opportunity to multiple graduate students and researchers. The students (from both plant science and engineering disciplines) gained valuable interdisciplinary expertise to successfully conduct cyber-agricultural research. They also had professional development experience via presenting their work in conferences and workshops. How have the results been disseminated to communities of interest?The results have been disseminated through the publications mentioned in the "products" section earlier. Also, the project directors and students have delivered talks in various relevant venues such as PAG and Phenome meetings. Our project "Artificial intelligence and crop scouting" is one of three picked projects to showcase Iowa State stories of statewide impact, innovation and entrepreneurship during the "2019 ISU Day" at the Iowa State Capitol on March 6, 2019. We also started the International Workshop Series on Machine Learning for Cyber-Agriculture Systems. We organized the second workshop of this series in ISU campus,Ames, Iowa. More than 120 researchers, students participated in the event from various academic institutions, companies and government agencies. What do you plan to do during the next reporting period to accomplish the goals?In the next reporting period (during the fourth no-cost extension year), we plan to do the following - We have collected UAV based RGB images of soybean diseases at three different heights (15m, 30m, 60m) and five different time points of plant growth during the 2018 and 2019 soybean growing seasons. However, we were unable to collect UAV hyperspectral images during the previous two growing seasons of the diseases due to technical issues (sensor was burnt due to overpower supply and had problem with continuous transfer of large hyperspectral data to onboard storage device during the flight). We plan to collect 125 band hyperspectral images (450-950nm) of soybean diseases (Iron Deficiency Chlorosis) using DJI M600 drone during the 2020 soybeans growing season. First, we will develop unsupervised segmentation algorithms for Soybean canopy segmentation from UAV images. We will then compare the performance between RGB and hyperspectral camera for the early disease identification. We will then develop a deep learning algorithm for soybean disease severity classification using the RGB and hyperspectral images. We will compare the disease classification accuracy of soybean diseases using the RGB and hyperspectral images captured at different height. We will develop an algorithm for UAV image super-resolution and will compare the disease identification performance between the images collected at 15m height and super-resolved images of 60m flight. UAV image super-resolution could enable us to cover large geographical area for disease detection by allowing us to fly at higher height without degradation in disease classification performance. We will develop phenotyping protocol (image frequency, Flight height, GCP's and data processing pipeline) for capturing UAV-based hyperspectral images. We will also develop an active learning algorithm to achieve high soybean stress classification performance under a fixed labeling budget. This will help us achieve the goals - Obj 4, 7 and 8.

Impacts
What was accomplished under these goals? In the thirdyear of the project, we focused on these following objectives - spatiotemporal data assimilation and predictive modeling for yield prediction, breeding support (Obj 3, 7), canopy-level stress identification in field using weakly supervised deep learning and active learning (Obj 3, 7) and robotic deployment of automated yield estimation (Obj 8).

Publications

• Type: Journal Articles Status: Published Year Published: 2020 Citation: Kevin G. Falk, Talukder Z. Jubery, Seyed V. Mirnezami, Kyle A. Parmley, Soumik Sarkar, Arti Singh, Baskar Ganapathysubramanian, and Asheesh K. Singh. Computer vision and machine learning enabled soybean root phenotyping pipeline. Plant Methods, 16, 5 (2020) doi:10.1186/s13007-019-0550-5
• Type: Journal Articles Status: Published Year Published: 2019 Citation: Kyle Parmley, Race Higgins, Baskar Ganapathysubramanian, Soumik Sarkar, and Asheesh Singh, Machine Learning Approach for Prescriptive Plant Breeding, Scientific Reports 9, Article number: 17132
• Type: Journal Articles Status: Published Year Published: 2019 Citation: K. Parmley, K. Nagasubramanian, S. Sarkar, B. Ganapathysubramanian, A. K. Singh, Development of Optimized Phenomic Predictors for Efficient Plant Breeding Decisions using Phenomics Assisted Selection in Soybean Plant Phenomics, Plant Phenomics (a Science Partner Journal), vol 2019, Article ID 5809404, 15 pages, 2019.
• Type: Journal Articles Status: Published Year Published: 2019 Citation: S. Ghosal, B. Zheng, S. C. Chapman, A. B. Potgieter, D. R. Jordan, X. Wang, A. K. Singh, A. Singh, M. Hirafuji, S. Ninomiya, B. Ganapathysubramanian, S. Sarkar, W. Guo, A weakly supervised deep learning framework for sorghum head detection and counting, Plant Phenomics (a Science Partner Journal), vol 2019, Article ID 1525874, 14 pages, 2019.

Progress 03/01/18 to 02/28/19

Outputs
Target Audience:During the secondyear of this project, we communicated our research outcomes to the broad scientific research community via multiple publications in reputed journals and conferences. Apart from this, we have also presented our work in the 2018NSF CPS PI meeting during the CPS Agriculture workshop. Our work has also reached the soybean farmers community (primarily in Iowa) and agriculture companies such as Monsanto, Corteva through various meetings with the project director and co-directors. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project has provided significant training and professional development opportunity to multiple graduate students and researchers. The students (from both plant science and engineering disciplines) gained valuable interdisciplinary expertise to successfully conduct cyber-agricultural research. They also had professional development experience via presenting their work in conferences and workshops. How have the results been disseminated to communities of interest?The results have been disseminated through the publications mentioned in the "products" section earlier. Also, the project directors have delivered talks in various relevant venues such as PD Sarkar delivered a Keynote address - Deep Learning for Agricultural Analytics, Keynote address in Asia-Pacific Federation for Information Technology in Agriculture/World Congress on Computers in Agriculture (AFITA/WCCA), IIT Bombay, Mumbai, India, October, 2018. Our team was also invited to the 2018 Farm Progress Showwhere we showcased our Soybean disease detection appwork using multi-stresses detection app to midwestern farmers along with international farmers from brazil, Argentina and China etc. as well as to the to US Secretary of Ag Mr. Sonny Purdue.Our project "Artificial intelligence and crop scouting" is one of three picked projects to showcase Iowa State stories of statewide impact, innovation and entrepreneurship during the "2019 ISU Day" at the Iowa State Capitol on March 6, 2019. We also started the International Workshop Series on Machine Learning for Cyber-Agriculture Systems. We also co-organized a CPS-Ag workshop within the 2018 NSF CPS PI meeting. . What do you plan to do during the next reporting period to accomplish the goals?In the next reporting year, we plan to primarily workon modeling disease spread dynamics (Obj 4) and stress mitigation planning strategies (Obj 6). Other than that, we will also complete the other tasks especially on the deploymentaspect.

Impacts
What was accomplished under these goals? In the first year of the project, we focused on these following objectives -Objective 2 - Create disease signature analytics using multi-modal data,Objective 3 - Development of a neural-symbolic framework for spatiotemporal data modeling, Objective 4 - Causal graphical modeling of disease spread dynamics,Objective 7 - Demonstrate the efficacy of the multiscale decision support framework on field scale, andObjective 8 - Buildinga large scale deployment framework. We made significant progress on all these objectives. We built deep learning based models for analyzing both RGB and Hyperspectral information for plant disease identification (Obj 2). We also built an mobile app for Soybean disease detection using deep learning.We made algorithmic developments in the area of neural-symbolic modeling,Granger-causal graphical modeling and deep LSTM models for spatiotemporal data modeling (Obj 3,4). For large field scale deployment, we built a label-efficient deep learning framework that can be used efficiently for UAV based images (Obj 7,8).

Publications

• Type: Journal Articles Status: Published Year Published: 2018 Citation: K. Nagasubramanian, S. Jones, S. Sarkar, A. K. Singh, A. Singh, B. Ganapathysubramanian, Hyperspectral band selection using genetic algorithm and support vector machines for early identification of charcoal rot disease in soybean stems, Plant Methods (2018)
• Type: Journal Articles Status: Published Year Published: 2018 Citation: A. K. Singh, B. Ganapathysubramanian, S. Sarkar, A. Singh, Deep learning for plant stress phenotyping: trends and future perspectives, Trends in Plant Science (2018)
• Type: Journal Articles Status: Published Year Published: 2018 Citation: A. Akintayo, G. Tylka, A. K. Singh, B. Ganapathysubramanian, A. Singh and S. Sarkar, A deep learning framework to discern and count microscopic nematode eggs, Scientific Reports 8, Article Number: 9145 (2018).
• Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Koushik Nagasubramanian, Sarah Jones, Asheesh Singh, Arti Singh, Soumik Sarkar and Baskar Ganapathysubramanian, "Hyperspectral Band Selection using Hybrid Machine Learning For early identification of Charcoal Rot Disease in Soybean Stems", AFITA/WCCA 2018.
• Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David Blystone, Sambuddha Ghosal, Asheesh Singh, Baskar Ganapathysubramanian, Arti Singh and Soumik Sarkar, "An Understandable DCNN Framework For Soybean Stress Phenotyping", AFITA/WCCA 2018.
• Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Johnathon Johnathon, Linjiang Wu, Tryambak Gangopadhyay, Baskar Ganapathysubramanian, Soumik Sarkar and Asheesh Singh, "Integrating genotype and weather variables for soybean yield prediction using deep learning", AFITA/WCCA 2018.
• Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Sambuddha Ghosal, Bangyou Zheng, Scott Chapman, Andries Potgieter, David Jordan, Xuemin Wang, Asheesh Singh, Arti Singh, Baskar Ganapathysubramanian, Soumik Sarkar and Wei Guo, "A Label Efficient Deep Learning Framework for Sorghum Head Detection and Counting", AFITA/WCCA 2018.

Progress 03/01/17 to 02/28/18

Outputs
Target Audience:During the first year of this project, we communicated our research outcomes to the broad scientific research community via multiple publications in reputed journals and conferences. Apart from this, we have also presented our work in the 2017 NSF CPS PI meeting during the CPS Agriculture panel and in the US-Japan collaboration meetings in Tokyo. Our work has also reached the soybean farmers community (primarily in Iowa) and agriculture companies such as Monsanto through various meetings with the project director and co-directors. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?The project has provided significant training and professional development opportunity to multiple graduate students and researchers. The students (from both plant science and engineering disciplines) gained valuable interdisciplinary expertise to successfully conduct cyber-agricultural research. They also had professional development experience via presenting their work in conferences and workshops. How have the results been disseminated to communities of interest?The results have been disseminated through the publications mentioned in the "products" section earlier. Also, the project directors have delivered talks in various relevant venues as mentioned in the "target audience" section earlier. What do you plan to do during the next reporting period to accomplish the goals?For the next reporting period, we are working on completing the Objectives 1, 2 and 5 that we have already made significant progress on. Apart from that, we have also started working on Objectives 3, 4 and 6.

Impacts
What was accomplished under these goals? In the first year of the project, we focused on these following objectives - Objective 1 - Building a layered sensing architecture, Objective 2 - Create disease signature analytics using multi-modal data?and Objective 5 - Hierarchical active information gathering. We have made significant progress in all three objectives. For Objective 1, we have set up hardware/software architecture and data gathering protocol for ground-based manual/robotic imaging and UAV based imaging. We have established a partnership with Planet Labs to gather relevant satellite data. For Objective 2, we have built and successfully demonstrated deep learning based multi-modal (RGB and Hyperspectral images) data analytics for plant disease detection. Finally, we have developed initial algorithms for optimally deciding the data necessary for row crop stress management.

Publications

• Type: Journal Articles Status: Published Year Published: 2017 Citation: Jubery, Talukder Z., et al. "Deploying Fourier coefficients to unravel soybean canopy diversity." Frontiers in plant science 7 (2017): 2066.
• Type: Journal Articles Status: Published Year Published: 2018 Citation: Ghosal, Sambuddha, et al. "An explainable deep machine vision framework for plant stress phenotyping." Proceedings of the National Academy of Sciences (2018): 201716999.
• Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: K. Nagasubramanian, S. Jones, A. K. Singh, A. Singh, B. Ganapathysubramanian, S. Sarkar, Explaining hyperspectral imaging based plant disease identification: 3D CNN and saliency map, Workshop on Intrepreting, Explaining and Visualizing Deep Learning...now what?, NIPS 2017.
• Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: S. Ghosal, D. Blystone, A. K. Singh, B. Ganapathysubramanian, A. Singh and S. Sarkar, Interpretable Deep Learning applied to Plant Stress Phenotyping, Symposium on Interpretable Machine Learning at NIPS 2017.