Deep computer vision system for cocoa classification

Abstract Cocoa hybridisation generates new varieties which are resistant to several plant diseases, but has individual chemical characteristics that affect chocolate production. Image analysis is a useful method for visual discrimination of cocoa beans, while deep learning (DL) emerged as the de facto technique image processing . However, these algorithms require large amount data and careful tuning hyperparameters. Since it necessary acquire number images encompass wide range agricultural products, in this paper, we compare Deep Computer Vision System (DCVS) traditional (CVS) classify beans into different varieties. For DCVS, used Resnet18 Resnet50 backbone, CVS, experimented machine algorithms, Support Vector Machine (SVM), Random Forest (RF). All were selected since they provide good classification performance their potential application food A dataset with 1,239 samples was evaluate both systems. The best accuracy 96.82% DCVS (ResNet 18), compared 85.71% obtained by CVS using SVM. essential handcrafted features reported discussed regarding influence on bean classification. Class Activation Maps applied DCVS’s predictions, providing meaningful visualisation most important regions model.