所謂的瑕疵豆就是指品質不好的咖啡豆，這些瑕疵豆會讓一杯咖啡產生令人不舒服 的味道，現在挑選瑕疵豆方法大多還是以人為方式為主，通過眼睛檢查進行挑選。然而 這種人工分類方式不僅耗費人力與時間，而且常常會因為人為的疏失以及不同人的判斷 標準不一致，造成篩選品質不穩定的問題。這些問題可藉由自動檢測機制來加以克服， 因此，本研究提出一個自動分類咖啡豆的模型，此模型以VGG16 為主架構，透過轉移 學習提升咖啡豆的分類準確度，實驗顯示本模型能得到99%的準確率。

The so-called defective coffee beans refer to coffee beans of poor quality. These blemish beans can give a cup of coffee an uncomfortable taste. At present, most of selecting defective beans is still artificially, and the selection is made through eye examination. However, this manual classification method not only consumes manpower and time, but also often causes the problem of unstable screening quality due to human negligence and inconsistent judgment standards of different people. These problems can be overcome by automatic detection mechanism. Therefore, this research proposes a model for automatically classifying coffee beans. This model uses VGG16 as the main architecture to improve the classification accuracy of coffee beans through transfer learning. Experimental results show that this model can achieve 99% accuracy.