Graduation Year

2023

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Electrical Engineering

Major Professor

Ismail Uysal, Ph.D.

Committee Member

Nasir Ghani, Ph.D.

Committee Member

Wilfrido Moreno, Ph.D.

Committee Member

Srinivas Katkoori, Ph.D.

Committee Member

Ultan McCarthy, Ph.D.

Keywords

Artificial Intelligence in Agriculture, Smart Supply Chain, Machine Learning, IIoT, Harvest Prediction

Abstract

The present dissertation embodies a multi-faceted investigation aimed at enhancing strawberry production pre-harvest and post-harvest quality through data-driven approaches. It unfolds multi-stages framework.

The first stage presents a complete data collection and its preliminary statistical analysis obtained from the first phase of a large-scale soil study on how to improve strawberry production and achieve sustainable and high-quality harvests through sensor-assisted real-time field monitoring. Six real-time loggers were placed in an operational commercial strawberry farm in Central Florida for the entirety of a harvest season from soil preparation to planting to harvesting. Along with high-resolution soil sensory measurements including water content, electrical conductivity, and temperature. Concurrently, strawberries were harvested from each of the six locations on three separate occasions for their objective physicochemical characteristics to be monitored and recorded in a food chemistry lab. The primary goal of this stage is to introduce the dataset to the food science and engineering research community and present the results of its preliminary statistical analysis to identify which factors correlate with one another. Based on the findings of Chapter 3, while there exists a weak correlation between the quality of the harvest and the water content of the soil immediately preceding it, there were several cases where statistically significant differences exist between the soil sensory measurements from different locations which did not replicate the same differences in their corresponding harvest qualities.

Chapter 4 outlines the second stage, as it describes an end-to-end system that integrates modern hardware and software tools for precise monitoring and control of soil conditions. In the proposed framework, the data from the initial stage is utilized to predict the physicochemical characteristics of the fruit at the point of harvest around the sensor locations. Empirical and statistical models are jointly investigated in the form of neural networks and Gaussian process regression models to predict the most significant physicochemical qualities of strawberries. Color, for instance, either by itself or when combined with the soluble solids content (sweetness), can be predicted within as little as 9% and 14% of their expected range of values, respectively.

The final phase is the post-harvest phase when the strawberry is harvested in the desired quality. The concept of using digital twins to represent a physical object or process in a digital environment, such as the cloud, has become more practical with the ubiquitous application of sensors and other IoT devices. Specifically for perishable items such as fresh produce, once the environmental data is collected, modern machine learning and AI algorithms can be used to create digital twins representing the instantaneous quality and marketability of the product to create unprecedented insights into the supply chain. For instance, the shelf-life of a strawberry can be predicted throughout the entire distribution process based on initial quality, temperature, and shipment duration. Strawberry marketability and quality including color, sugar content (sweetness) and firmness deteriorate post-harvest, and accurate prediction of these metrics can aid in maintaining quality standards during distribution and enable smart supply chain. In Chapter 5, we introduce an AI/ML driven digital-twin for strawberries and demonstrate that using simple sensor data, we can create a reliable representation of the marketability index, color, sugar content, and firmness of a digital strawberry with prediction error percentages of 4.21%, 15.56%, 1.95%, and 4.12% within the expected range of values, respectively. The support vector machine was used for regression in the first two models, while decision tree regressions were used for the second two models. Ultimately, the goal is to enable first-expired-first-out distribution to replace the industry standard first-in-first-out, which creates preventable loss with the use of accurate and validated digital twins.

Collectively, this research marks the first integration of sensor-assisted real-time field monitoring, advanced predictive modeling, and AI-driven digital twins for strawberries. It provides both the academic and industrial communities with a comprehensive, data-centric approach to improving strawberry production and supply chain management. Through this work, the research offers a groundbreaking pathway for sustainable and high-quality strawberry production and distribution, harnessing the unparalleled potential of digital technologies and revolutionizing current agricultural practices.

Download

Share

COinS