Implementation of MobileNetV2 Transfer Learning for Image-Based Classification of Cocoa Fruit Diseases

Abstract

Cocoa is one of the high-value agricultural commodities in Indonesia; however, its productivity continues to decline due to the increasing prevalence of plant diseases, particularly Black Pod Disease and attacks by Helopeltis spp. Traditional disease detection, which is generally performed manually by farmers, is often inefficient and prone to errors, thereby highlighting the need for an intelligent and technology-assisted early diagnosis system. This study aims to develop a disease classification model for cocoa plants using a Convolutional Neural Network (CNN) based on the MobileNet-V2 architecture, which is recognized for its computational efficiency and strong performance in image analysis. The dataset consisted of 300 images divided into three categories: healthy cocoa pods, Black Pod Disease, and damage caused by Helopeltis. Following the Pareto principle, 80% of the data were allocated for training and 20% for testing. The model was trained for 10 epochs with a batch size of 32 and was supported by data augmentation to improve data variability. Experimental results demonstrated a significant improvement in performance, with the highest validation accuracy of 93.75% achieved at the seventh epoch. The confusion matrix further confirmed that the model classified each category with a high level of precision. These findings indicate that MobileNet-V2 is an effective approach for automatic cocoa disease detection and has strong potential to assist farmers in improving disease management practices in the field.Key words: Convolutional Neural Network; MobileNet-V2; Cocoa Disease Classification; Helopeltis spp.; Deep Learning