Study Of Two-dimensional Crystals Growth Based On Real-time Image Analysis

Industrial crystallization technology has a wide range of applications in the fine chemicals and pharmaceutical industries.In order to obtain the desired crystal shape and ensure the quality of the crystal product,it is necessary to monitor the crystal growth process in real time for realtime control and optimization.In recent years,more and more image monitoring technologies have been applied to perform real-time online analysis of industrial crystallization processes.This paper firstly introduces the application status and research progress of image detection technology in industrial crystallization process,and analyzes the advantages of realtime image system detection of crystallization process and the problems in practical application.For needle-shaped crystals,an image processing method suitable for in-situ image detection can overcome the adverse effects caused by uneven illumination,particle motion,and solution agitation,eliminate noise information in images to improve image quality,and enhance crystal shape regions in images.The image processing method mainly includes image enhancement processing based on multi-scale Retinex and image segmentation processing based on minimum cross entropy,thereby effectively extracting a crystal shape region from the original image.In order to monitor the growth quality of crystal population online during the crystallization of L-glutamic acid,an image analysis method based on non-invasive microscopic imaging system for real-time detection of crystal population growth rate was proposed.The image segmentation algorithm based on Mask Regional Convolutional Neural Networks(Mask R-CNN)is used to analyze the in-situ images in real time based on the offline sample training weight network model.The algorithm can accurately classify and segment overlapping and non-overlapping crystals in the image acquisition process,and generate a pixel-level precision target mask.Accordingly,the crystal two-dimensional size measurement is performed for each crystal image according to the ellipse fitting.Then,the Gaussian kernel function based density estimation is used to smooth the numerically calculated Crystal Size Distribution(CSD)for visual evaluation.The size distribution and shape distribution of the crystal population can be monitored in real time by introducing two indicators describing the properties of the crystal population,namely the Symmetric Variant of Relative Entropy(SVRE)and the CSD dispersion.And a method for calculating the crystal population growth rate in real time is given.A real-time image monitoring interface was designed for the needle crystal growth process.The effectiveness and practicability of the proposed image analysis method was verified by real-time monitoring experiments on the cooling crystallization process of ?-L-glutamic acid.