Sack Detection and Counting Using Deep Learning

Abstract

In this paper, a grain sack detection and counting system is developed to help the logistics management of a warehouse stock. As main objective is to construct an electronic device that performs the counting of sacks with great precision through the most advanced artificial vision techniques. The device will count the total number of sacks that make up a stowage, calculate the volume, and eventually estimate the amount of mass, these results are transferred to an Excel sheet for constant monitoring and easy handling for users. The detection model is carried out with Python, PyTorch and YOLOv3, obtaining as a result a mean Average Precision (mPA) of 0.92. In addition, a sacks stowage arrangement was established that allows the volume result to be as accurate as possible, likewise, a program that performs the calculation was developed.