Snacks Detection Under Overlapped Conditions Using Computer Vision

Abstract

This research aims to detect and classify snacks. The detection and classification process uses the Mask R-CNN algorithm. The training process is carried out in the training stage with 250 epochs and 150 steps per epoch. The dataset used in this study consists of 687 snack images with a resolution of 640 x640 pixels divided into 549 training data and 137 validation data. In addition, System testing results were conducted using scenarios 1-7 in an overlapping or partially covered state within the 10-70% range. It can be interpreted that snack overlap detection has optimal performance in the 10-50% range, as evidenced by the high mAP value of 0.99. However, the system cannot detect well in the 60% and 70% overlap range, as seen from the low mAP values of only 0.2 and 0. The evaluation results show that the system has an excellent performance in performing object detection and classification tasks with high accuracy and consistency.