Discrimination of Cocoa Beans using Structural Image Features: An Experimental Analysis

This paper proposes a machine vision solution for classification of cocoa beans using structural characteristics like shape and size of the cocoa beans extracted from images using classic machine learning techniques. The features are extracted from images through a chain of image processing techniques.Finally traditional machine learning techniques (KNN, SVM, Decision Tree and Random Forest) are employed to classify the cocoa beans in four classes e.g. large, medium, small and rejected. Comparative studies among different techniques are also performed. Prior to train the model, extracted features are optimized by two optimization techniques-Univariate Selection and Feature Importance. Trained models are evaluated using stratified K-fold cross validation and finally mean cross validation scores are calculated for performance analysis. The experimental results show that the Random Forest Classifier provides highest accuracy score of 0.75.

I. Introduction

Cocoa is the key ingredient in chocolate and chocolate confections. Chocolate is incredibly popular, and it is one of the common foods in the world. About 300 to 600 cocoa beans are required to make 1 kg of chocolate, but this may vary depending on the cocoa content in the beans. The cocoa nibs, cocoa paste (mass or liquor), butter, powder and couverture etc. are the byproducts of the cocoa beans. These are the main ingredients to make chocolate and other food products. The beans are also used to make cosmetics and soaps. Cocoa beans are produced in tropical zones around the Equator. About 70 percent of the world’s cocoa beans come from West African countries likeGhana, Nigeria Ivory Coast, Indonesia and Cameroon.