我国陕南寒武纪初期的宽川铺组(距今大约535百万年)以盛产磷酸盐化的微体小壳化石和球状动物胚胎化石而闻名于世，是探索早期动物门类起源和寒武纪生命大爆发研究中的一个非常重要的窗口。在宽川铺组中寻找微体化石新类型主要依赖显微镜下人工挑选的传统方法日趋低效，迫切需要新的技术手段来解决这些问题。而微体化石数量庞大，大小接近，形态简单，非常适合于采用机器学习的手段进行人工智能分类。我们在宽川铺组微体化石中尝试使用机器学习的手段来进行化石图像识别和机器分拣。采用方向梯度直方图来提取化石图像的主要向量特征，并设计了二叉树型多分类识别器进行化石数字分类。目前处理了5 000多张微体化石照片的人工智能识别，已经取得了较高的准确率。

The early Cambrian Kuachuanpu Formation in south Shaanxi, China is well known for phosphatized small shelly fossils and metazoan embryo fossils, thus it is a critical window for investigating the origin of metazoans and Cambrian explosion. Finding the valuable specimens among numerous microfossils is quite low in efficiency based on the manually selecting under the microscope, thus we developed a method of machine learning and artificial intelligence for recognition and classification of microfossils. We have acquired more than 5 000 images of major taxa from the Kuanchuanpu Formation and reached a high precision rate in image processing and fossil recognition.

国家自然科学基金 (项目编号：41672009，41621003，41772010，41720104002)；“中国科学院战略性先导科技专项 (B 类 )(编号XDB26000000)”共同资助