Implementation of Faster R-CNN with Colour and Blur Augmentation For Differentiate Cloves From Debris

Identification of cloves for debris separation (leaves and twigs) is very important in improving the quality of clove processing in the industry. This research develops a computer vision-based clove and debris object detection system using the Faster R-CNN algorithm with the ResNet50 architecture as an object detection model. The dataset consists of images of cloves and debris consisting of leaves and twigs, sometimes involved in weighing the cloves. This research utilizes the process of color augmentation and blurring to manage images during training as a contribution to this research. The testing phase was carried out using several schemes, namely comparing the detection results from testing using Faster RCNN without treatment and Faster R-CNN using color augmentation and blur. The experimental results show that the developed model can detect objects with better accuracy than the Faster R-CNN model without treatment, with mAP values of 87% and 97%, respectively. In addition, this study also evaluates the resulting model by comparing the test results with differences in speed and density resulting from data collection using a conveyor belt. The best accuracy of 96% is obtained at a minor density with a speed of 60 rpm.