Performing Department
(N/A)
Non Technical Summary
Addressing food waste is crucial for sustainability, especially as the global population grows. From growers to consumers, inefficiencies lead to significant food waste. Consumers prefer visually appealing produce, prompting growers and packers to discard a significant percentage of their product due to misalignment between supply and demand. Efficient models that integrate real-time data can enhance agricultural operations by improving order fulfillment and inventory selection, thereby reducing waste and optimizing harvest, storage, and packing processes.Small and medium-sized growers are particularly affected by food loss and waste, which add financial pressure and can deter sustainable farming practices, making growers more vulnerable in the market. Consequently, these growers risk being edged out, reducing the diversity of available produce and increasing centralization in the agriculture sector.To address the gap between supply and demand, this proposal aims to quantify the impact of inventory management and high-throughput grading technologies on food waste for small- and medium-sized growers selling produce through consignment sellers. The central hypothesis is that improved knowledge of produce grade, both just-in-time and historical, will decrease waste and increase profitability for small- and medium-sized growers. This hypothesis will be tested through three main objectives:1) Develop and implement systems to quantify produce waste: Install camera systems at Nash Produce to monitor the pickout and eliminator table belts, producing metrics related to grade, weight, and produce characteristics. This will inform dashboard metrics estimating packout, profitability, revenue, and efficiency.2) Develop an online learning system for produce order-inventory matching: Update inventory grade distribution to optimize order matching as data becomes available; and3) Create models and methods for defining post-harvest initial conditions: Develop a sweetpotato growth model using various degrees of prior knowledge. Validate these models through packing line experiments.The focus will be on sweetpotato as a high-value horticultural crop representative of many graded crops. The results will be broadly applicable to consignment sellers of any graded crops, enhancing sustainability and profitability across the agricultural sector.
Animal Health Component
40%
Research Effort Categories
Basic
20%
Applied
40%
Developmental
40%
Goals / Objectives
Objective 1: Develop and implement systems to quantify produce waste. Install camera systems at a consignment packer to monitor both the pickout belt and the eliminator table belt, enabling metrics production related to grade, weight, and characteristics of produce. This will inform dashboard metrics estimating packout, profitability, revenue, and efficiency.Objective 2: Develop an online learning system for produce order-inventory matching. As data becomes available, the system will update inventory grade distribution to optimize order matching as more is learned about each field.Objective 3: Create models and methods for defining post-harvest initial conditions. To estimate grade, develop a sweetpotato growth model using varying degrees of a priori knowledge. Data sourcing will be from publicly available repositories. Validate models using experiments on the packing line.
Project Methods
Objective 1: Develop and Implement Systems to Quantify Produce WasteModern agricultural supply chains are complex, often relying on growers' and packers' experience rather than data. To address this, systems will be developed to provide real-time insights into produce condition. Camera systems at a commercial consignment seller will monitor the pickout belt and eliminator table to create metrics on grade, weight, and produce characteristics. Key research questions include the accuracy of grading algorithms and the effectiveness of collected metrics in quantifying production efficiency.Task 1.1 involves installing cameras at a commercial production facility that operates as a consignment seller, replicating a system used at another facility. Smaller edge devices will collect data on pickout waste streams and communicate wirelessly to a main GPU-enabled device.Task 1.2 will develop produce grading models using algorithms to determine produce quality based on surface and shape defects. These metrics will be integrated into a dashboard to inform packout value estimation.Milestones include the successful installation of camera systems and the incorporation of new metrics into the dashboard.Objective 2: Develop Online Learning System for Order-Inventory MatchingAgricultural consignment sellers need to adapt operations based on dynamic demand and inventory conditions. This objective focuses on creating analytical tools for order-inventory matching, leveraging phenotypic data, statistical modeling, and predictive analytics to reduce waste and optimize turnover rates. Key research questions explore the system's decision-making capacity with varying data points and the impact of scanning materials before packing.Task 2.1 involves collecting and integrating real-time data from multiple sources, expanding a knowledge graph to connect relevant identifiers to each sweetpotato instance.Task 2.2 will develop adaptive algorithms to estimate inventory grade distributions, using Gaussian distributions initially and refining with real-time data. Reinforcement Learning and Bayesian updating will enhance the system's accuracy.Task 2.3 focuses on optimizing order matching and inventory selection using in-silico estimations and a Monte Carlo Tree Search algorithm.Milestones include completing adaptive algorithms and integrating a functional order-matching system into the dashboard.Objective 3: Create Models for Defining Post-Harvest Initial ConditionsConsignment sellers rely on historical information and employees' mental models to predict inventory grade. This objective aims to develop models for grade prediction based on field and growing conditions and create tools for employees to input estimates. Key research questions address the effectiveness of mental models, the minimum variables needed for accurate predictions, and the impact of combining different predictors.Task 3.1 will elicit mental models from employees using Decision Intelligence to create a Causal Decision Diagram (CDD) and translate these models into computer models.Task 3.2 will refine an existing sweetpotato yield prediction model using varying levels of information, from basic geographical data to detailed soil and nutrient data.Task 3.3 will validate predictive models by comparing them against actual data from the eliminator table, assessing profitability and accuracy.Milestones include developing initial predictive models and validating them experimentally.Quantifying the Impact on Small- and Medium-Sized FarmsThe impact of the system on small- and medium-sized growers will be evaluated over years 2 and 3, focusing on profitability changes.Experiment 1 will establish a baseline by monitoring the consignment sellers' current practices and simulating optimal packout processes to assess potential waste reduction and profit impacts.Experiment 2 will evaluate the decision support tool's effectiveness in real packing scenarios, ensuring stakeholder satisfaction and achieving significant waste reduction.Milestones include validating the system through simulations and real-world testing, demonstrating statistically significant improvements in waste reduction and profitability.