LocAnalyzer: A computer vision method to count locules in tomato fruits

Abstract

Fruit production represents an important economic resource for many countries. The size and shape of fruits are crucial in production systems because they affect the market value. Particularly, in tomatoes (Solanum lycopersicum), the number of seed-containing cavities within each fruit, called locules, affects these attributes. The number of locules is also relevant for genotype-phenotype association studies designed to accelerate tomato breeding programs. Traditionally, the determination of the number of locules was performed through a visual inspection of a cross section of the fruit, a laborious, time-consuming, and highly subjective task. In this work, an automatic computer vision method for the identification and counting of the number of locules in tomato fruit images, called LocAnalyzer, is presented. The aim of LocAnalyzer is to speed up the processing time and reduce the subjectivity relative to the traditional manual approach for locule counting. The method was tested on two real tomato datasets. Promising results in terms of accuracy, precision, and recall measures were obtained, suggesting the potential usefulness of the approach for the development of a tool for the automatic measurement of other internal tomato attributes. Additionally, an experiment comparing the capacity of domain experts and LocAnalyzer in the identification of the number of locules was performed. The fact that expert-LocAnalyzer inter-group dispersion is smaller than expert-expert dispersion suggests that LocAnalyzer could be used as a gold standard for counting the number of locules. The proposed method, which is freely available, was implemented in the R programming language, and a web-based application was developed for online tests.