Abstract
For online monitoring of crystal population growth quality during the L-glutamic acid crystallization process, an efficient image analysis method is proposed to detect the crystal population growth rate in real time, based on a non-invasive imaging system. A trained weights network model is established for real-time analysis of in-situ captured images, by using a deep-learning based image segmentation algorithm named Mask R-CNN (Regional Convolutional Neural Network), which is capable of accurate image segmentation and classification of crystals in blurred images. Correspondingly, crystal size measurements are conducted for each crystal image in terms of elliptical fitting. Then, density estimation with Gaussian kernel function is used to smooth the numerically computed crystal size distribution (CSD) for evaluation. By introducing two indices for describing crystal population properties, namely symmetric variant of relative entropy (SVR) and CSD dispersion, a real-time computation method is given to estimate the crystal population growth rate. Experimental results on the cooling crystallization of $\beta$ form L-glutamic acid well demonstrate the effectiveness of the proposed image analysis method.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have