Texture-Based Classification of Cocoa Pod Diseases and Pests Using GLCM and LDA on Field-Acquired Dataset

Abstract

Cocoa pod diseases and pests are major factors that reduce cocoa yield quality and productivity. This study proposes a texture-based image classification approach for identifying cocoa pod conditions, including diseases and pest damage, using the Gray Level Co-Occurrence Matrix (GLCM) for feature extraction and Linear Discriminant Analysis (LDA) for classification. The original dataset, consisting of 420 field-acquired images across four classes (blackpod, healthy, helopeltis, and tupai), was expanded through augmentation to 1,260 images to improve model robustness. Model performance was evaluated using two complementary evaluation schemes: hold-out testing with active data augmentation and stratified k-fold cross validation. Experimental results show that the hold-out evaluation achieved an accuracy of 82.11%, indicating good practical performance when trained on augmented data. Meanwhile, stratified 5-fold cross validation produced an average accuracy of 77.70% with a standard deviation of 4.49%, providing a more conservative estimation of model generalization under limited original data conditions. The results demonstrate that classification performance is strongly influenced by the evaluation strategy and data availability. Overall, the combination of GLCM and LDA proves effective for texture-based classification of cocoa pod diseases and pests and can be implemented in a web-based application to support practical usage. However, further improvements in robustness and generalization are expected through the inclusion of more diverse original images and additional feature representations.