Detection and Classification of Defective Fresh Excelsa Beans Using Mask R-CNN Algorithm
Analyn N. Yumang, Ma. Chloe M. Sta. Juana, Regina Liza C. Diloy
Abstract
This study focuses on creating a system that detects and classifies defective fresh Excelsa beans. Mask R-CNN is a Convolutional Neural Network model that predicts classes and generates bounding boxes and segmentation masks for each class. The pre-trained model that will be used is called Detectron2 Mask R-CNN, and it will be trained in the Google Colab to learn the features of each Defective Fresh Excelsa bean, namely, Black Bean, Sour Bean, Cut Bean, and Insect Damaged Bean. To test the model’s accuracy in detecting and classifying the model, the researchers will use the Raspberry Pi 4 with a camera and take a picture of 40 fresh Excelsa beans. The system will automatically detect and classify the Fresh Excelsa bean. Then the output will be displayed in the Raspberry Pi LCD. With the gathered data, the model achieved an accuracy of 87.5%.