Improving Classification Accuracy of Local Coconut Fruits with Image Augmentation and Deep Learning Algorithm Convolutional Neural Networks (CNN)

Abstract

Local coconut varieties must be classified to maintain the quality and genetic diversity of coconuts as the main commodity in Indonesia's largest coconut-producing region. This study introduces a deep learning module for improved classification of coconuts, using color jitter as part of a data augmentation strategy to supplement the existing dataset and utilizing well-known CNN-based models like VGG16 for image analysis, with a focus on the needs of future research. The goal is to improve the classification accuracy of local coconut varieties through deep learning. We investigate both data augmentations and EDA, and we use VGG-16-based CNN models to enhance the classification performance. We used a confusion matrix for the model evaluation, containing metrics like accuracy, precision, recall, and f1-score. Results reveal that a color jitter augmentation model attained a training accuracy of 99.12%, testing accuracy of 97.33%, and validation accuracy of 97.33%. Model exploration using VGG16, on the other hand, improved all three: training accuracy—99.87%, testing accuracy—98.77%, and validation accuracy—98.97% average F1-score: 99%. Our research contributes massively to providing the best automatic classification method that will benefit and help farmers shorten their jobs while promoting economic growth in trading effectively across Indonesian regions. Its novelty is in combining image augmentation and CNNs, concerning the VGG16 model, showing better.