应用卷积神经网络分类的Hindeodus牙形刺细粒度数据集

随着人工智能浪潮的兴起,利用卷积神经网络对化石进行分类识别已得到越来越多的关注,并表现出了巨大的应用潜力。通过调研发现,前人所分类的化石物种基本属于不同的属、科或更高级的生物分类单位。然而,现实中对于同属异种化石的鉴定往往是重点和难点,也意味着前人所训练的分类器可能并不能很好地用于实际的化石鉴定。鉴于此,本文通过文献收集,建立了一个包含12种同属于Hindeodus属的牙形刺数据集,同时提供了对原始数据增强后的数据集。由于该数据集具有细粒度的特点,用户可以使用卷积神经网络并结合细粒度图像特征提取技术对其进行训练。针对数据集存在数据量较少、类别不均衡等不足,建议用户在训练时使用分层K折交叉验证、迁移学习和加权损失函数等手段来解决以上问题。本文数据集旨在为生物化石智能识别领域增补一个细粒度化石数据集,其可作为卷积神经网络对细粒度(种一级)化石进行智能鉴定的实验数据集。本数据集所遵循的细粒度原亦可以作为建立其他门类化石数据集的参考。

Comment on“First Record of Hindeodus-Isarcicella Population in Lower Triassic of Slovenia”by Tea Kolar-Jurkovek and Bogdan Jurkovek in Palaeogeography,Palaeoclimatology,Palaeoecology,252:72-81

Some conodonts described by Tea Kolar-Jurkovsek and Bogdan Jurkovsek （2007） from the Permian-Triassic boundary section in the Ziri area of western Slovenia are revised in this paper. One specimen that they assigned to Hindeodus parvus is tentatively regarded as Hindeodus cf. scalaris Wu 2005. Two specimens that they assigned to lsarcicella sp. A is changed to Hindeodus parvus. One specimen that they described as lsarcicella lobata Perri and Farabegoli is changed to Hindeodus crenatus Wu （2006）. One specimen that they assigned to Isarcicella turgida （Kozur, Mostler and Rahimi-Yazd） is changed to Hindeodus cf. parvus. Since these revisions, the three conodont zones that they defined have been correspondingly revised. Their first zone may belong to Hindeodus parvus zone, but it is not certain. Their second zone is changed to the Hindeodus parvus zone, and their third one remains as is.

浅析P/T界线层中牙形刺Hindeodus parvus的标准性

本文讨论了中国(江西修水东岭、四川广元上寺和浙江长兴煤山)、印度(克什米尔Guryul Ravine)、前苏联(阿塞拜疆Achura)等地的海相P/T界线层中牙形刺Hindeodus parvus,及其首次出现的点位,阐述了其标准性对海相P/T界线划分的重要意义。建议以产于单相且沉积连续剖面,主齿和细齿均保存完好的Hindeodus parvus的首次出现的点位,作为P/T生物地层界线划分的标准。