Automated Detection of Coffee Bean Defects using Multi-Deep Learning Models

Traditional methods of coffee bean inspection mainly rely on manual labor, which results in issues such as low efficiency and subjectivity. However, with the advancement of computer vision and machine learning technologies, automated coffee bean inspection has become possible. Therefore, this study aims to develop a system that can distinguish between good and bad coffee beans using two main components: YOLOv7 and convolutional neural network (CNN). The image recognition model in this study is divided into three categories: broken, insect-infested, and mold, all of which employ transfer learning. Using YOLOv7, the coffee beans are easily recognized and processed by the image classification model Then, the captured coffee beans are used as the input for the image classification model. If the output results are all negative, it means that the bean is good, and it will be kept. However, if there is at least one output indicating a defect, the bean will be labeled as the corresponding defect type. In the end, using the DenseNet201 model, we achieve an accuracy of 98.97% in classify defective coffee beans.