<p>Identifying coffee bean varieties and roasting levels is crucial for promoting Saudi coffee culture. Varieties like Harari, Khawlani, Nabari, Laqamti, and Bariah can be distinguished by their unique color, shape, and morphological traits. Similarly, roasting levels, from "green" to "dark roast," are identified by color. Traditional visual and mechanical inspection methods have been inadequate for accurately distinguishing these varieties and roast levels. To address this, we present a deep learning model that predicts coffee type and roast level from images. We curated a "Saudi Coffee Type" dataset with 4 roast levels and 5 Saudi coffee bean types. Our model, based on the SqueezeNet architecture with Vision Transformer (ViT) enhancements, effectively interprets this dataset. It uses a multi-head attention mechanism, achieving 85.9% accuracy, an F1 score of 0.871, and an F2 score of 0.872. These metrics demonstrate the model's capability to predict coffee type and roasting level, understanding bean differences. Comparative analyses show our model excels in classifying Saudi coffee beans by morphological traits and accurately categorizing roasting levels through color analysis. Future research will aim to generalize this approach for broader applications, potentially extending it to other global coffee varieties. Enhancements will include expanding the dataset, refining the model architecture, and collaborating with coffee experts. Our model represents a significant advancement in coffee classification, leveraging machine learning to support Saudi coffee culture preservation and promotion.</p>