近年来，人工智能的兴起推动了我国煤炭行业在绿色环保、智能生产等方面的发展。针对芳香晶格条纹识别过程中存在的解译慢、周期长和误判率高等问题，提出了一种基于MASK R-CNN的智能提取方法。该智能提取方法通过调整特征参数和优化网络结构，对条纹特征的识别算法进行改进:① 跳过前期预处理过程，以端到端的模式进行芳香晶格条纹的特征提取训练，实现从原始HRTEM中自动识别芳香晶格条纹的功能;② 突破常规神经网络提取目标物外轮廓的思维桎梏，实现对线要素的快速准确分割。以人工解译结果为标准，通过智能提取结果和传统提取结果在精度评价、识别效果和参数提取等方面的对比，验证该智能提取方法的有效性。结果表明，在识别精度方面，智能提取方法的准确率、精准率、召回率和交并比的数值为91.2%，85.2%，83.4%和72.9%，分别高于传统提取方法的89.9%，62.1%，75.4%和51.6%;在识别效果方面，智能提取方法对细节的处理更加智能，提取的线条连贯性和平滑性更好，同时噪音更少;在长度和取向等参数的统计分析方面，智能提取结果与人工解译结果在取向和长度分布图上的匹配度更高。综上所述，该智能提取方法通过MASK R-CNN进行芳香晶格条纹的自动识别，降低工作成本，提高工作效率，为煤大分子结构研究提供更快捷可靠的数据支持。

In recent years,the rise of artificial intelligence has promoted the development of green environ mental protection and intelligent production in China coal industry.To solve the problems of slow interpretation,long period,and high misjudgment rate in the recognition of aromatic lattice fringes,an intelligent extraction method based on MASK R CNN was proposed.The intelligent extraction method improves the recognition algorithm of lattice fringes by adjusting the feature parameters and optimizing the network structure:① skipping the prepare process,the extraction training of aromatic lattice fringes in end to end mode was performed and the function of automatically identifying aromatic lattice fringes from the original HRTEM was realized;and ② Breaking through the mental shackles of conventional neural networks to extract the outline of the target object,the line elements was split quickly and accurately.Taking manual interpretation results as the standard,the effectiveness of the intelligent extraction method was verified by comparing the results of intelligent extraction with traditional extraction results in terms of accuracy evaluation,recognition effect,and parameter extraction.The results show that in terms of recognition accuracy,the ac curacy,recall rate,and cross to merge ratio of the intelligent extraction method reach up to 91.2%,85.2%,83.4%,and 72.9%,separately,while for those of traditional extraction method,the values are 89.9% and 62.1%,75.4%,and 51.6%,respectively.In terms of recognition effect,the intelligent extraction method has more intelligent ability in the detailed processing,and has more consistency and smoothness in the extracted lines,and also has less noise than traditional extraction method.In terms of statistical analysis in the length and orientation distribution,intelligent extraction and manual interpretation show a high degree of matching.Therefore,the intelligent extraction method based on MASK R CNN can automatically identify aromatic lattice fringes,which can reduce operating cost and improve working efficiency,and this can provide a reliable support for studying the coal macromolecular structure.