基于两级分段式算法的卷积神经网络的浮游有孔虫自动鉴定*

有孔虫个体微小、数量众多、地理分布广、演化迅速, 是记录海洋沉积环境的重要载体, 在海相生物地层划分和对比中具有十分重要的作用。因有孔虫属种众多, 传统的属种鉴定需要经验丰富的专业人员进行人工鉴定且耗时较长, 此外人工鉴定古生物面临人才匮乏和工作量大等问题。卷积神经网络在计算机视觉领域的应用可较好的解决上述问题。利用古生物专家对中新世浮游有孔虫化石标注为指导, 根据有孔虫化石不同方向的视角分类, 结合卷积神经网络算法, 开发了有孔虫化石图像识别系统。研究发现, 通过有孔虫化石腹视、缘视和背视角度分类, 采取两级分段式鉴定算法对中新世浮游有孔虫属一级进行识别, 属一级鉴定准确率达到82%左右。

Automatic identification of planktonic foraminifera using convolutional neural networks based on two stage-segmentation algorithm

Foraminifera are characterized by their small size, abundance, wide distribution, and quick evolution. They provide important information about the sedimentary environment in the ocean and are significant for biostratigraphic division and correlation in marine facies. Traditional identification of foraminifera requires experts with rich experiences and is very time consuming. Fossil identification by human faces lots of problems such as the lack of talents, the amount of work, and communication difficulties. Convolution neural network algorithm, having been widely applied in computer visual identification field, can solve these problems effectively. Guided by labeled Miocene planktonic foraminifera, we develop an image recognition system by integrating different visual angles of the fossils and convolution neural network algorithm. The results indicate that, by applying the algorithm of two stage-segmentation, the fossils can be identified at generic level using the images of the umbilical, spiral, and edge views of the specimens. The identification accuracy of the computer model at generic level for the Miocene planktonic foraminifera can reach about 82%.