The research of ostracod fossils is of importance to the determination of geological age, the study of paleo- oceans, the reconstruction of paleo- environments and the exploration of submarine oil resources. However, the existing methods for identifying fossil are time- consuming and labor- intensive, and the accuracy rate needs to be improved. In this paper, we propose a hierarchical recognition method due to the hierarchical structure of ostracod fossils’ categories(families, genera, and species) and wide range of species.Methods:First, perform object detection to realize the positioning and genus- level classification; then intelligently identification based on the object detection module, uses CNN to extract features among same genus particle and SVM for species- level classification.Results:It is demonstrated that the hierarchical identification can locate and classify particles in fossil images and our test presents the 95% classification accuracy. The accuracy increases by 1.8%~5.8% which are compared with non- hierarchical recognition one.Conclusions: In this paper, the accuracy of the proposed method reaches 95%, which confirms the feasibility and application prospects of computer vision methods based on deep learning in paleontology research. The computer directly gets the paleontological image feature through learning and classify automatically, making full use of the computer's active learning characteristics. In the future research work, we will further expand the ostracod fossil database, the data samples and categories to improve the model of recognition accuracy, generalization and applicability. Based on the proposed model, an intelligent ostracod fossils identification system will be developed to improve the efficiency of ostracod fossil identification.