Image-Based Crystal Size Analysis for β- form L-Glutamic Acid Crystallization Via Deep Learning-Based Object Detection

For evaluating the quality properties of p-form L-glutamic acid (LGA) crystals during a crystallization process, an image analysis method with deep learning-based object detection is proposed to acquire the crystal classification and size measurement, based on the optimization of in-situ image quality. Firstly, the homomorphic filtering is adopted to enhance the contrast of snapshotted images and highlight the image contours of crystals. Secondly, the crystal images are classified according to the degree of blur, and meanwhile, YOLOv6 is utilized for model training to obtain the position of individual crystals in a snapshotted image. Thirdly, the size of non-adhesive crystal image with clearness is calculated by a boundary fitting algorithm to improve the measurement accuracy. The effectiveness of the proposed image analysis method is verified by numerical simulation and experiments on the β-form LGA crystallization process.