Convolutional Neural Network to enhance stock taking

Abstract

In the current competitive economic environment, there is a lot of focus in optimization of processes and providing high quality customer experience. This study explores the use of deep learning particularly convolutional neural network to enhance the retail store stock taking process. It provides a review of literature on different convolutional neural network architectures to identify the best fit for image object detection and count. It highlights some of the image analysis applications in various sectors such as counting fish, yield estimation and construction site management. YOLO is noted to be perform well based on the literature review and the study further implements and compares the performance of YOLO v2 and YOLO v3 in object detection and count. The implementation leverages on the pretrained weights on ImageNet and further training is done on open data image set of retails stores. Both YOLOv2 and YOLOv3 achieve mean average precision above 75%, however YOLOv3 is leading attaining a mean average precision of 81.86%. Keywords convolutional networks, YOLO, stock take