Ethiopia's primary export is coffee, which contributes significantly to the country's foreign exchange earnings. In Ethiopia, there are numerous types of coffee beans. Based on their respective geographic origins, these beans differ from one another in terms of quality, color, shape, and other characteristics. These coffee beans are categorized based on where they were grown. However, visual inspection, which is subjective, time-consuming, and prone to error, is used to categorize coffee beans. The goal of this research was to create a suitable model that categorizes various Ethiopian coffee varietals according to the location where they were grown. Six coffee-growing regions provided sample coffees (Harar, Jimma, Limu, Sidama, Yirgachafe, and Wombera). To reduce environmental influences and eliminate variance, an Epson L360 scanner was used to capture input images. 100 images were taken from each region. The total number of images taken was 600, which contain 12604 coffee beans. Out of the total data sets, 75% were utilized for training and 25% were used for testing to assess the classification accuracy. Morphological operations and marker-based watershed algorithms were used to segment coffee beans even if they were mutually touched. In this research, to address the limitations of CNN on texture and shape feature extraction, we have used GLCM and the regionprops function. Sixty four CNN features, five morphological features, and five GLCM features were extracted from each coffee bean for the classification process. KNN and SVM classifiers have been evaluated on the extracted feature. The experimental result showed that combining CNN, morphology, and GLCM features gave better classification accuracy. And also, the SVM classifier achieved 90.16 percent accuracy on the combined feature. Keywords: Ethiopian coffee bean, CNN, GLCM, KNN, SVM.