过去的学者从地球化学分析结合岩相学观察来探讨峨眉山玄武岩高钛低钛类型成因，至今尚无定论。利用晶体粒度分布分析方法来研究玄武岩结构的特征从而判别其成岩所经历的岩浆作用动力过程，尽管被证实有效，但因为操作技术的问题，其在统计学上的可靠性有待商榷。本项目基于深度学习神经网络相关技术方法，改进从岩石薄片图像中自动抓取晶体粒度分布数据的方式，提高数据获取的效率，使得大量数据的快速获取成为可能；建立监督学习下的晶体粒度分布大数据的神经网络分类模型，对数据特征进行智能分类从而识别其所对应的岩浆作用过程，解决人为分析的低效问题、提高分析的科学性，让晶体粒度分布分析符合大数据及相关分析技术；并将改良后的分析技术应用在判别攀西地区峨眉山玄武岩成岩过程中的岩浆动力作用，结合钛含量，在之前地球化学及岩相学的研究基础上，厘清高钛低钛玄武岩类型的成岩成因，从而为钒钛磁铁矿和铜镍硫化物矿床的成矿成因提供理论指导。

Although lot of geochemical studies as well as petrographical observations have been paid much attention to Emeishan flood basalt in the past decades worldwide, its high-Ti and low-Ti type petrogenesis is still poorly determined. Crystal size distribution (CSD) analysis can be explored to research the basalts textures quantitatively so as to determine the petrological processes. However, due to the difficulty to extract CSD data from thin sections largely and quickly, CSD analysis is weak robustness in statistics. In this proposal, we will utilize deep learning algorithms (e.g. convolutional neural network) to develop a new way of extract CSD data from thin sections quickly and largely. Furthermore, we will construct a neural network model of multi-class classification to detect patterns of basalts textures so as to intelligently determine the drive forces of petrological processes, avoiding low efficiency through manual analysis in the past and enhancing the analytical scientificity. The advanced CSD analysis technique will be applied to judge the petrological processes of Emeishan flood basalts in Panxi district, together with Ti contents, so as to ably understand the petrogenesis of high-Ti and low-Ti types spatiotemporally associated with magmatic Fe-Ti oxide deposits and Ni-Cu-(PGE) sulfide deposits, and help clarify the disagreements by past geochemical studies.