Method For Online Measurement Crystal Morphology In Cooling Crystallization Process Based On In-situ Microscopic Image

The cooling crystallization process is an important part of obtaining solid crystal products from a reaction solution containing one or more components,and is widely used in fine chemical and pharmaceutical engineering.In order to optimize the cooling and crystallization process in real time,there is an urgent need for advanced online detection instruments and technologies for monitoring the crystal growth process.In recent years,the use of in-situ microscopic images to analyze the crystal morphology and population size distribution in the crystallization process has received more and more attention and discussion.This article takes the L-glutamic acid cooling crystallization experiment as a research case,studies the crystal growth size measurement method based on microscopic image detection,and designs an online monitoring software and hardware platform.First introduce the crystallization process in-situ microscopic image detection device and experimental design scheme used in this article,and analyze the characteristics and difficulties of the microscopic crystalline image compared with the macroscopic object image.Subsequently,the basic steps of crystal image segmentation are explained.For the case where the image is clear and complete,and the segmentation requirements are not high,a monocular image segmentation method is given,and the image segmentation algorithm is used to perform rapid segmentation.For more complex situations,a deep learning network based on Mask-RCNN The image segmentation method can achieve better crystal image segmentation effect,and design a special crystalset file format to compress and save the image segmentation result.Secondly,for a large number of low-quality crystal images in the segmentation results,a reference-free image quality evaluation algorithm is proposed,which can dynamically filter out high-quality images.Perform matching analysis on the binocular collected images to distinguish overlapping crystals for crystal size measurement and distribution statistics.Considering the obvious difference in image characteristics at different stages of the cooling and crystallization process,an image detection cooling and crystallization process stage division method is proposed in order to achieve automatic switching of the image detection model and algorithm in stages,so as to ensure reliable crystal population size distribution statistics.In order to facilitate engineering application,design a crystallization process online monitoring platform program.From the bottom layer to the top layer,it is divided into four layers: hardware layer,middle layer,algorithm layer and application layer.The lower layer provides support for the upper layer,and the upper layer calls and controls the lower layer.The hardware layer contains on-site image acquisition and preprocessing system and network communication structure.For the middle layer,a data flow graph programming framework is designed.The outstanding advantage is that it is convenient for users to combine any functional sub-modules through graphical drag-and-drop connections to build a parallel accelerated computing structure.Each sub-function is written as a plug-in function node for calling.And design a crystal analysis data structure to provide efficient,flexible and lightweight data storage.