Enhanced Classification of Coffee Bean Varieties in the Philippine Market using MobileNetV1

Classifying coffee bean varieties is essential for quality control and productivity in the coffee industry. Traditional manual classification methods are labor-intensive, error-prone, and inefficient, particularly in markets with diverse coffee varieties like the Philippines. While existing research has explored deep learning approaches for this task, lightweight architectures such as MobileNetV1, designed for resource-constrained environments, remain largely underutilized. This study addresses this gap by developing a MobileNetV1-based model and benchmarking its performance against prior efforts using a dataset of 12,000 images of four locally sourced coffee varieties: Arabica, Excelsa, Liberica, and Robusta. The proposed model achieved an overall accuracy of 94.33%, surpassing the highest benchmark by 4.33%. It delivered near-perfect accuracy for Liberica (99%) and Arabica (98%), alongside competitive performance for Robusta (93%) and Excelsa (88%). These results highlight the potential of lightweight deep learning models to achieve both scalability and accuracy in agricultural classification tasks, particularly in resource-limited regions. By demonstrating MobileNetV1’s ability to enhance classification accuracy while reducing computational demands, this study offers a practical, cost-effective solution for automating coffee bean classification, paving the way for wider adoption of AI in precision agriculture.