Abstract In the process of chemical production and biopharmaceutical, with the complex reaction, the products will overlap or adhere to each other. Effectively distinguishing the overlap and adhesion of crystals is of great significance for the statistics of different morphological characteristics such as the number and size of crystals. This paper proposes a crystal texture recognition system based on image analysis, which mainly includes image pre-processing, feature extraction and texture classification. Firstly, the crystal images are pre-processed to eliminate the influence of water droplets, particle shadows and uneven illumination. Secondly, the Improved-Basic Gray Level Aura matrix (I-BGLAM) is used to extract texture features of the crystals to determine the focus state of crystals. Finally, the texture features are classified by back propagation neural network (BPNN) to effectively distinguish agglomerates and pseudo-agglomerates. The case study and experimental results of cooling crystallization of l-glutamic acid show that the texture recognition system can effectively distinguish the adhesion and overlap of crystals, and effectively analyze the agglomerates, and has good experimental accuracy.

中文翻译：

---

基于图像分析的晶体纹理识别系统用于分析团聚体

摘要 在化工生产和生物制药过程中，由于反应复杂，产物会相互重叠或粘连。有效区分晶体的重叠和粘附，对于晶体数量、大小等不同形态特征的统计具有重要意义。本文提出了一种基于图像分析的晶体纹理识别系统，主要包括图像预处理、特征提取和纹理分类。首先对晶体图像进行预处理，以消除水滴、粒子阴影和光照不均匀的影响。其次，利用改进的基本灰度光环矩阵（I-BGLAM）提取晶体的纹理特征来确定晶体的聚焦状态。最后，通过反向传播神经网络（BPNN）对纹理特征进行分类，以有效区分团聚体和伪团聚体。L-谷氨酸冷却结晶的案例研究和实验结果表明，纹理识别系统可以有效区分晶体的粘附和重叠，有效地分析团聚体，具有良好的实验准确性。