Computer Vision system for rock classification using Artificial Neural Network

Abstract

This project has been proposed by the Nijst company, located in Belgium. This company works with natural stone tiles and cobblestones. One of its activities is the purchase of large quantities of old road stones at very low prices, however the problem is that these stones, come in batches with different types mixed. At the moment, there are two operators in charge of the manual sorting, and the objectives are to reduce sorting time and costs. This project explores the design of a computer vision system that is able to classify between the different types of cobblestones on a conveyor belt and separate them into different boxes. To carry out the recognition part, a computer vision system with a camera and lighting inside a black box, was designed. Additionally, a python script based on the OpenCV library was written, to detect the cobblestone and track it. After that, a Artificial Neural Network (ANN) was trained with histogram features such as, weighted mean, skewness, and kurtosis to classify the different types of cobblestones. Finally, for the separation part, a script was made with PLC programming so that when the rock passes next to its corresponding box, an actuator would push it inside. Overall, it is concluded that due to the high classification accuracy, the correct performance of the system, and the extraordinary profitability, this project that has been carried out has been a success, heralding the promising future of artificial neural networks for classification tasks.