Transfer Learning on Semantic Segmentation for Sugar Crystal Analysis

Abstract

In sugar factories, crystal particle analysis plays an important role in the quality of sugar production. Analyzes include measuring the crystals' dimensions, estimating the area and distribution of the crystals in terms of dimensions, which are of great help in setting up sugar kilns. To be able to analyze sugar particles, we must first be able to segment crystals correctly. Therefore, segmentation is the first and most important stage of the analysis. This paper introduces a method based on Transfer Learning (TL) in deep neural networks for the Semantic Segmentation of sugar crystals. In this method, by modifying a pre-trained Convolutional Neural Network (CNN) called DeepLab, the semantic Segmentation of sugar crystals is performed and the results clearly show that this method can label the crystals with high accuracy and remove extra parts.