Classification of Cocoa Pod Maturity Using Similarity Tools on an Image Database: Comparison of Feature Extractors and Color Spaces

Abstract

Côte d’Ivoire, the world’s largest cocoa producer, faces the challenge of quality production. Immature or overripe pods cannot produce quality cocoa beans, resulting in losses and an unprofitable harvest. To help farmer cooperatives determine the maturity of cocoa pods in time, our study evaluates the use of automation tools based on similarity measures. Although standard techniques, such as visual inspection and weighing, are commonly used to identify the maturity of cocoa pods, the use of automation tools based on similarity measures can improve the efficiency and accuracy of this process. We set up a database of cocoa pod images and used two feature extractors: one based on convolutional neural networks (CNN), in particular, MobileNet, and the other based on texture analysis using a gray-level co-occurrence matrix (GLCM). We evaluated the impact of different color spaces and feature extraction methods on our database. We used mathematical similarity measurement tools, such as the Euclidean distance, correlation distance, and chi-square distance, to classify cocoa pod images. Our experiments showed that the chi-square distance measurement offered the best accuracy, with a score of 99.61%, when we used GLCM as a feature extractor and the Lab color space. Using automation tools based on similarity measures can improve the efficiency and accuracy of cocoa pod maturity determination. The results of our experiments prove that the chi-square distance is the most appropriate measure of similarity for this task.

Classification of Cocoa Beans by Analyzing Spectral Measurements Using Machine Learning and Genetic Algorithm

“…Machine learning is a subfield of artificial intelligence that focuses on creating models and algorithms to learn from data and make predictions or classifications without explicit programming. In our work [14], we employed the following algorithms:…”

Analysis of the Impact of Color Spaces on Skin Cancer Diagnosis Using Deep Learning Techniques

“…Each color space has its specific advantages and uses in medical imaging and other areas related to computer vision [22].…”