Pengembangan model klasifikasi kualitas sarang burung walet berbasis CNN dengan transfer learning MobileNetV2

Abstrak

The quality of edible bird’s nests (EBN) is a crucial factor in determining their market value, which necessitates an accurate, automated classification system. This study aims to develop a quality classification model for EBN using a Convolutional Neural Network (CNN) approach with transfer learning based on the MobileNetV2 architecture. The dataset consists of 3,406 EBN images collected directly from farmers, processed through aggressive data augmentation and background removal to emphasize the main object. The data were evenly split into training (2,723 images) and validation (683 images) sets, covering three quality classes: high, medium, and low. The model was trained in two phases: initial training with frozen base layers, followed by fine-tuning. Evaluation results showed an increase in validation accuracy from 93% to 97% after fine-tuning, with average precision, recall, and F1-score values of 0.97. The confusion matrix indicated high classification consistency, with most predictions aligned along the diagonal. This study contributes to the development of a high-accuracy image-based EBN quality classification system by integrating contour-based preprocessing, data augmentation, and CNN architecture optimization. The resulting model offers a reliable automated solution suitable for industrial implementation to support objective and efficient EBN quality sorting