Instance Segmentation and Classification of Coffee Leaf Plant using Mask RCNN and Transfer Learning

Abstract

Coffee is one of the most consumed beverages in the world. It is crucial in the economy of many industrial companies in developing countries. This study proposes a deep learning algorithm called Mask RCNN to segment coffee leaves from complex real-world backgrounds and classify them as healthy and unhealthy. The RoCole dataset was manually labeled using the VGG Image annotator. The algorithm uses Resnet101 and the FPN architecture for feature extraction. The RPN creates region proposals for each feature map to separate the input image from the background. The system has a high-test accuracy of 97.76% for the binary classifier. If the image is classified as unhealthy, it goes through another segmentation stage based on the HSV color model to highlight the defective areas of the coffee leaf. The instance segmentation results showed that the mAP@50:95 was 100%, the recall@50:95 was 84.5%, and the F1-score was 91.6%.