Comparison of Deep Learning Technologies Applied to the Recognition of Defects in Cocoa Beans

Cocoa beans play an essential role in global agriculture, serving as the foundation for the beloved chocolate production, therefore, their quality is essential. The conventional manual quality classification process can be time consuming. This study proposed a quality control technique using machine learning models, during which it was delved into cocoa bean production processes, neural networks, and machine learning algorithms. In this experiment, two distinct approaches were explored; the first method involved the use of extracted features from cocoa bean images for the training and evaluation process of several machine learning models. The second method involved the use of a deep learning algorithm directly fed with labeled images to predict the quality status of the cocoa bean. For the extracted features method, we evaluated various machine learning models, including a decision tree classifier, a random forest classifier, a support vector machine, and a artificial hydrocarbon network. For the deep learning approach, roboflow was used as a tool to train and assess the model, the labeled images were directly imported to train the model. Through comprehensive training and evaluation, the results of each method were compared with each other and previous studies. Interestingly, the deep learning model implemented with roboflow demonstrated superior performance, which proved to be a highly viable method for controlling cocoa bean quality.