Progress 07/01/24 to 06/30/25

Outputs
Target Audience:During this reporting period, our efforts have directly reached two primary target audiences: consignment sellers (packers) and the small- and medium-sized growers they serve. We also engaged with the broader agricultural science community. Our most intensive engagement was with our consignment packing partner, Nash Produce. At their facility, we successfully reached managers, IT staff, and packing line operators through the hands-on, in-person installation and testing of project hardware and software. This included deploying cameras and sensors on the packing line to measure produce input, waste, and output (Objective 1). We also developed and deployed a production dashboard that displays real-time eliminator table weights and counts, along with a novel data-drift tool that compares incoming produce with order requirements (Objective 2). These efforts provided the packing house staff with immediate, actionable data to inform operational decisions. Furthermore, we conducted a decision elicitation session with their expert staff to capture the "mental models" used to match inventory to orders, ensuring our future models are grounded in practical expertise (Objective 3). While our work was physically situated at the packing house, the ultimate beneficiary and a key target audience is the community of small- and medium-sized growers. The data collected and the tools developed directly pertain to the quality, packout, and subsequent profitability of their sweetpotatoes. The systems for quantifying waste and optimizing order matching are designed to increase the portion of their crops sold, directly impacting their revenue. By improving the packing facility's ability to estimate and manage post-harvest quality, we provide an indirect but key service to these growers who rely on a consignment seller. Finally, our work in developing and validating novel yield quality prediction models (Objective 3) and data-driven decision support tools (Objective 2) contributes directly to the agricultural science and agri-tech communities. The methodologies and findings from this project will be disseminated through academic channels, providing new insights into reducing food waste through technology and predictive analytics. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?This project has provided significant opportunities for training and professional development for a postdoctoral researcher, two graduate students, a research scholar, and our industry partners. These activities have fostered a cross-disciplinary environment, enhancing skills in engineering, data science, and agricultural technology. For the Postdoctoral Researcher and Research Scholar: The project supported the professional development of one postdoctoral researcher and one research scholar by tasking them with creating key software components. The postdoc led the development of the "production dashboard" and inventory management tool, gaining valuable experience in full-stack development, UI/UX design for an industrial setting, and translating stakeholder needs into a functional, data-driven software product. The research scholar developed a novel "data drift" tool, a professional development activity that increased their knowledge in applied data science, algorithm development, and statistical analysis for real-time process control. For Graduate Students: The project has provided direct, hands-on training and professional development for two graduate students from different engineering disciplines: An Electrical and Computer Engineering (ECE) graduate student received one-on-one training and mentorship in applying engineering theory to a practical, real-world challenge. They gained significant professional skills by leading the design, construction, and installation of the camera and sensor systems. Their development was further enhanced by managing the on-site edge computing devices and coordinating the complex data ingress from the hardware to the cloud platform, fostering critical project management and collaboration skills. An Agricultural Engineering graduate student has been trained in grower-focused data collection and management methodologies. Their professional development has involved working to source and curate diverse datasets related to growing conditions (e.g., planting dates, locations, grower practices). This work has provided them with a unique, cross-disciplinary skill set, bridging the gap between agronomic field data and the data requirements for building and validating predictive yield quality models. For Industry Partners: The project also created professional development opportunities for the managers and staff at Nash Produce. Through collaboration on the decision elicitation session and the deployment of the production dashboard, they were introduced to new data analytics tools and concepts. This training provided them with new skills to interpret real-time data, helping to integrate data-driven decision-making into their daily packing operations. How have the results been disseminated to communities of interest?Results and project progress have been disseminated directly and continuously to our primary communities of interest through two main channels: Monthly Production Team Meetings: We hold regular meetings with our industry partner, Nash Produce, which include their operational managers and our research team. In these sessions, we present project updates, demonstrate newly developed tools like the production dashboard and data drift feature, and review the initial data being collected from the sensors. This collaborative forum allows for immediate feedback, ensures our work remains aligned with their operational needs, and serves as the primary method for deploying and training on the project's outputs. Meetings with Consignment Growers: We have engaged directly with the small- and medium-sized growers who operate under Nash. In these meetings, we disseminate the project's goals, share how the technologies being implemented at the packing house aim to increase their profitability by reducing waste, and explain how data from their produce contributes to the project's success. This direct engagement keeps the primary beneficiaries informed and ensures the project's outcomes will be understood and valued by the grower community. What do you plan to do during the next reporting period to accomplish the goals?Develop a Pipeline to Ingress Track & Trace Bintag Data (Objective 1):Bin tags will be read and data ingressed into the dashboard system automatically after the bins are dumped into the washtank. This will potentially enable the key track and trace metrics -- such as grower name, field, harvest date, etc. -- to be automatically associated with the imagery collected from the Eliminator Table sensor. Finalize the Online Learning System (Objective 2): The immediate future work involves completing the final task for the online learning system: "Optimal Inventory Selection" and finalizing the adaptive algorithms. This phase focuses on fully integrating the adaptive algorithms with historical and incoming order data (pending Nash's inventory management software integration) to create a robust tool that provides actionable recommendations for matching inventory to orders. The completion of this task in the current quarter (Y2 Q2) will mark the end of the primary development phase for this objective. Refine and Validate Post-Harvest Models (Objective 3):We will continue to develop the yield prediction models and their subsequent"Iterative Validation." From the current quarter through the end of the project, we will continuously validate and refine the models by comparing their predictions of produce quality against the real-world data being collected from the packing line sensors. This long-term process is critical for improving the models' accuracy and reliability, but will first require full integration of our data collection into one of our collaborating consignment farms' John Deere Operations Center. Conduct Major System-Wide Experiments: The final phase of the project involves two major experiments to test the system's real-world impact: Baseline and In-Silico Validation: Starting in(Y2 Q2) and continuing into early next year (Y3 Q1), we will conduct simulated experiments. Using historical data, we will test how the integrated system--combining the yield prediction models and the order optimization tools--would have performed, establishing a baseline for its potential impact on waste reduction and profitability. Live Online Testing: Beginning at the end of this year (Y2 Q4) and continuing through most of Year 3, we will deploy the validated system for live online testing at the packing facility. This will involve using the tool's recommendations to guide real-world packing decisions. We will measure the system's performance against traditional methods to quantify its effectiveness in a live operational environment. Finally, we will work to produce papers related to inventory matching and automated methods of training classification algorithms based on the waste stream camera.

Impacts
What was accomplished under these goals? During this reporting period, significant progress was made across all three project objectives. Objective 1: Develop and implement systems to quantify produce waste. We have successfully achieved most of the milestones for this objective, moving from design to implementation. The team completed the design, parts ordering, and construction of the primary sensor system for the eliminator table. This integrated sensor system was successfully installed on the packing line at Nash Produce in November. Following installation, we established the complete data management pipeline. This involved enabling remote access for software testing and updates, configuring data ingress into both Globus and Azure cloud platforms, and initiating software-scheduled continuous data collection. To capture additional data points, cameras have also been successfully installed over the waste stream and at the dumptank to scan bin tags for traceability, although the software pipeline for the bin tag scanner is still pending. Objective 2: Develop an online learning system for produce order-inventory matching. Foundation work for the online learning system is complete. A robust data pipeline and database were created specifically for the data flowing from the newly installed eliminator table sensors. To make this data actionable, we developed and deployed a "Production dashboard" for use by the packing facility, which visualizes key metrics like real-time weights and counts. A key decision-support tool, the "Data drift" feature, was also completed. This tool provides a direct, quantitative comparison of the size and weight distribution of incoming produce against the specifications of uploaded customer orders. The development of more advanced order-matching tools is pending the integration of data from the Nash Produce's inventory management software system. Objective 3: Create models and methods for defining post-harvest initial conditions. Progress has been made on the foundational and modeling tasks for this objective. We successfully completed the first task by conducting a "Decision elicitation session" with packing house experts. This session captured the crucial "mental models" and decision-making processes used to match produce inventory to orders, providing a qualitative framework for our quantitative models. Work is now in progress on the second task: refining and expanding the sweetpotato yield prediction model. This effort involves enhancing the model to predict yield quality based on varying levels of available pre-harvest data, from basic location and date information to more detailed agronomic data. The model is currently being validated against the real-world data now being collected from the eliminator table.

Publications