Dynamic crystal monitoring method for crystallization process based on deep learning

Abstract

The real-time monitoring of the crystallization process plays an important role in the optimal control of the crystallization process. Because of the characteristics of multiple concurrences and irregular morphology in the crystallization process, this paper starts from the two aspects of target detection and image segmentation to explore the actual and effective dynamic monitoring method in the crystallization process. By using YOLOV5s target detection algorithm to detect the crystal position in the image to determine the target position. In addition, the adaptive threshold segmentation method and Sobel edge feature extraction algorithm are used to achieve accurate positioning and segmentation of crystal target, to achieve the position detection of crystal in the image and accurate segmentation of edge information.