Convolutional Neural Network Architectures for Classification of Coffee Varieties in the Philippines

This study investigates the application of advanced deep learning architectures in classifying coffee varieties, addressing the complexities and demands of the global coffee market. Given the limitations of traditional classification methods, which are labor-intensive and error-prone, this research explores the potential of Convolutional Neural Networks (CNNs) to enhance the accuracy and efficiency of coffee variety classification. Specifically, the study evaluated five CNN architectures—VGG19, ResNet152v2, DenseNet201, NASNetMobile, and NASNetLarge—on a self-captured dataset of 12,000 images representing four major coffee varieties available in the Philippine market: Arabica, Excelsa, Liberica, and Robusta. The research employs an experimental design to systematically assess each model's performance in handling complex image data. The models are rigorously tested across standardized metrics, including accuracy, precision, recall, and F1-score, to determine their suitability for practical deployment in agricultural settings. Based on the results, DenseNet201 demonstrated the highest overall performance with precision, recall, and F1-score, all at 91% and an accuracy of 90%, suggesting its superior feature utilization and propagation capability. In contrast, other models like ResNet152V2 showed lower performance, indicating challenges in handling the dataset's variability. This study contributes to the technological advancements in agricultural practices and underscores the importance of integrating cutting-edge AI technologies to meet the specific needs of rapidly expanding markets like the Philippines. By providing a comparative analysis of CNN architectures, the research offers valuable insights into their application in precision agriculture, aiming to support sustainable development goals through improved crop management and quality assurance.