Computer Science Students Probe Tomato Attributes at Artificial Intelligence Conference

Research focuses on classification of popular garden fruit based on 34 morphological attributes

WOOSTER, Ohio — Some say tomāto; some say tomăto, but for seven computer science majors at The College of Wooster, it’s less about pronunciation and more about physical attributes — specifically the size, shape, and color of America’s popular garden fruit.

The students recently returned from Cincinnati, where they joined Sofia Visa, assistant professor of computer science at Wooster, in presenting their research about these variables at the 22nd annual Midwest Artificial Intelligence and Cognitive Science Conference (MAICS).

What, you might ask, do tomatoes have to do with artificial intelligence? More than you would think, according to Visa. “Using machine learning algorithms (algorithms that enable computers to learn rules from empirical data in much the same way that humans learn — from examples and through repetition), we train the computer to categorize the tomato fruits based on various morphological attributes,” she said. “The larger problem we are studying, together with the VanDer Knaap Laboratory at the Ohio Agricultural Research and Development Center (OARDC), is the identification of genes that affect particular shapes (e.g. elongated or round) and sizes (e.g. cherry versus beef) in the tomato fruits.”

As part of their machine intelligence class, the students engaged in research that focused on the classification of tomato fruits based on 34 morphological attributes (width, length, etc.). “We investigated several classification techniques and attribute-ranking methods on 416 samples of tomatoes distributed in eight classes (round, flat, heart, etc.)” said Visa. “Our goal was to find the best classification algorithm (or ensemble of algorithms) for this particular data set, and further to identify dependencies between the morphological and genetic data.”

The students comprised three different research teams, each of which contributed to the paper published in the conference proceedings. In addition, each team made poster presentations at the conference. The first group, which included Wooster’s Matthew Lambert and Benjamin Snyder, presented a poster on “Attribute Selection and Classification for Tomato Fruit Data - That's a RELIEF!” This project focused on ranking the 34 attributes based on their individual classification abilities and on identifying subsets of attributes that classify more accurately the tomato’s morphological data than when using all 34 attributes. The second group, which was made up of Joshua Thomas, Jacob Haning, and Yanlong Hu, presented a poster titled, “Feed Forward: Artificial Neural Networks Applied to Tomato Data Classification,” while the third group, with Michael Janning and Mohammad Ahmad, explained “Learning Morphological Data of Tomato Fruits — Analysis of Variance and the Nearest Neighbors.”

In addition, a fourth group of Wooster students participated with a paper and presentation at the conference. The group, which included Thomas as well as Itai Njanji and Atticus Jack, worked with OARDC collaborators last summer to develop new features and improved methods for assessing the morphology and color of digitalized images of tomato fruits available in the Tomato Analyzer 3.0 software. "Their charge was to improve the software's effectiveness in defining boundaries (the edge of each slice of tomato fruit) and detecting the color of the samples."

“This conference is usually for graduate students so it is very impressive that our undergraduates were able to present their research at this level,” said Visa.” This project is an important opportunity for integrating research into the classroom and showing the students that the theory they learn in the classroom can be applied to solve real-world problems.”

The project is funded by a National Science Foundation grant awarded to the OARDC in collaboration with The College of Wooster and Boyce Thompson Institute.