CN: 11-6035/N
ISSN: 2096-2223
其他数据论文 II 区论文（已发表） • 版本 ZH3 Vol 8 (2) 2023
下载
应用卷积神经网络分类的Hindeodus牙形刺细粒度数据集
A dataset of fine-grained fossils of the conodont genus Hindeodus for classification using convolutional neural networks
段雄
DUAN Xiong
 >>
： 2022 - 09 - 08
： 2023 - 03 - 22
： 2022 - 12 - 02
： 2023 - 04 - 06
Baidu
map
6251 19 0
摘要&关键词
摘要：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。
关键词：卷积神经网络；牙形刺；Hindeodus；细粒度
Abstract & Keywords
Abstract: With the rise of artificial intelligence, the booming application of convolutional neural networks to the classification and identification of fossils has attracted more and more attention. According to our survey, it is found that the species classified by previous authors basically belong to different genera, families or higher biological taxonomic units. However, in fact, the identification of fossils between species within a genus is often the focus and challenge for the identification task, which means that the previously trained classifiers may not be suitable for actual fossil identification. On this basis, in this paper, we built a dataset covering 12 species of the conodont genus Hindeodus by means of literature collection, while providing an augmented dataset of the original data. Since the dataset is fine-grained, users can train it by using convolutional neural network combined with fine-grained image feature extraction technology. In view of the deficiencies of the dataset such as small amount of data and unbalanced classes, it is suggested that users use stratified K-fold cross-validation, transfer learning and weighted loss function in the training task to solve the above problems. The dataset is aimed to add a fine-grained fossil dataset to the field of intelligent identification of biological fossils, which can be used as an experimental dataset for intelligent identification of fine-grained (species-level) fossils by convolutional neural networks. The fine-grained primitive followed by this dataset can also be used as a reference for the establishment of other fossil datasets.
Keywords: convolutional neural network; conodont; Hindeodus; fine-grained
数据库（集）基本信息简介
数据库（集）名称Hindeodus 牙形刺数据集
数据通信作者段雄（duanxiong00@163.com）
数据作者段雄
数据时间范围牙形刺数据沉积时间为晚二叠世至早三叠世
地理区域全球
数据量298.97 MB
数据格式*.jpg
数据服务系统网址http://doi.org/10.57760/sciencedb.j00001.00641
基金项目四川省自然科学基金（2022NSFSC1177）；西华师范大学博士科研启动项目（20E031）
数据库（集）组成数据集共包含3个子数据集（图像数据）和一个数据来源的附录文件（Word文档）。其中：(1)子数据集Raw_Dataset.rar是由原始数据构成，包含Hindeodus的12个种，大小为53.32 MB，共853个图像数据样本；(2)子数据集Raw_train_val_Dataset是由Raw_Dataset划分出训练集（80%）和验证集（20%）得到的；(3)Augmented_Dataset是由对Raw_train_val_Dataset中训练集进行数据增强后得到的，大小为192.33 MB，共计7693个图像数据样本；(4) 牙形刺化石图像源于60篇已公开发表的文献，参见附录文件。
Dataset Profile
TitleA dataset of the conodont genus Hindeodus
Data corresponding authorDUAN Xiong (duanxiong00@163.com)
Data author(s)DUAN Xiong
Time rangeFrom the Late Permian to the Early Triassic
Geographical scopeWorldwide
Data volume298.97 MB
Data format*.jpg
Data service systemhttp://doi.org/10.57760/sciencedb.j00001.00641
Source(s) of fundingNatural Science Foundation of Sichuan Province (2022NSFSC1177); Doctoral Research Project of China West Normal University (20E031).
Dataset compositionThe dataset contains a total of three sub-datasets (image data) and an appendix file (Word document) of data sources. Among them, (1) the sub-dataset Raw_Dataset.rar is composed of raw data covering 12 species of Hindeodus with a size of 53.32 MB and a total of 853 image data samples; (2) the sub-dataset Raw_train_ val_Dataset is obtained by dividing the training set (80%) and validation set (20%) from Raw_Dataset; (3) Augmented_Dataset is obtained by data augmentation of the training set in Raw_train_val_Dataset, with a size of 192.33 MB and a total of 7,693 image data samples; (4) The conodont images are derived from 60 published papers (see Appendix file).
Baidu
引 言
牙形刺是一种沉积在寒武系—三叠系海相地层中的磷灰质古生物化石，其中许多属种是用于地层划分和对比的标准化石，亦在沉积环境演化、解释盆地历史、区域变质研究和石油勘探等方面扮演着重要角色[1-2]。鉴定牙形刺的传统流程一般为：针对某一属种的牙形刺，由精通该属种的专家通过观察其形态和构造、查询相关文献及对照已有的化石标本、图版和描述资料等，并结合自身经验最终确定其归属。该过程通常会过于依赖特定分类群专家的先验知识，不仅费时耗力，且结果往往受到鉴定人员的专业水平和主观意识的影响。对于一些从事地球科学研究但没有古生物学专业背景的学者和专家而言，化石难以准确鉴定可能成为滞缓其研究进展的因素之一。
随着地质标本图像数字化程度的日益加深，极大地增进和拓宽了大数据挖掘和机器学习在地学领域的使用场景[3]，计算机视觉在古生物学中的潜在应用也备受关注。大量研究表明，机器能够接近人类对图像的识别能力，并且随着计算能力和视觉数据的增加，它可以与人类鉴别能力相媲美，不久的将来其效率和准确率会越来越高[4]。据此，结合日臻丰富的机器学习和深度学习理论知识，寻找一种智能且精确的化石鉴定手段不仅可以大大简化传统的化石鉴定过程，而且可以为非古生物学专业的科研学者提供鉴定化石的便捷途径，这对于广大地球科学研究人员具有重要的理论和实际意义。
近年来，得益于GPU算力的显著提升和深度学习的迅速发展，利用深度卷积神经网络（Convolutional Neural Network, CNN）对古生物化石（如牙形刺、有孔虫、腕足类、介形类、孢粉等）进行自动识别取得了丰硕的成果并展现出了巨大的应用潜力[5-12]。需要指出的是，以上研究基本是针对化石大类或者属间进行识别，而研究人员通常是针对某一段地层进行研究，故需鉴定的化石样本必然集中在生物大类的多个属或者属内多个种，然而对属内种间化石进行鉴别往往是工作的重点和难点。因此，笔者认为关于古生物化石智能鉴定中使用的数据集样本最好均为某类化石的同一个属，样本间关系为同属异种，且在数据搜集的过程中尽量覆盖所有种，以便最大程度地提高模型的实用性。鉴于以上考虑，本文已搜集牙形刺Hindeodus常见的12个种，构建了细粒度图像数据集用于模型训练。相较于前人所采用的中—粗粒度图像分类（图1a，图1b），由于这些牙形刺之间较为相似（图1c），故建议使用可提取细粒度图像特征的CNN（如双线性CNN [13]，嵌入注意力机制模块[14]的CNN）来捕获更加细微的差异，从而提高分类器的性能。


图1   不同粒度图像分类示意图
Figure 1 Schematic diagram of different grained classification
1   数据来源和处理方法
1.1   数据来源
本数据集所有Hindeodus牙形刺化石样本扫描电镜图像均来自近20年来国内外公开发表文献（见附录参考文献），所囊括的12个种分别是H. anterodentatus，H. bicuspidatus，H. changxingensis，H. eurypyge，H. inflatus，H. julfensis，H. latidentatus，H. parvus，H. postparvus，H. praeparvus，H. sosioensis，H. typicalis。同一个样本一般提供了口视、反口视和侧视三种视角的图像，本数据集只选取了最能展现其特征的侧视照片，同时剔除了部分化石严重不完整的图像，最终共计853个原始图像数据样本（表1）。
表1   用于分类的牙形刺Hindeodus数据集
种名数量种名数量
H. anterodentatus20H. latidentatus64
H. bicuspidatus20H. parvus266
H. changxingensis47H. postparvus25
H. eurypyge63H. praeparvus169
H. inflatus36H. sosioensis13
H. julfensis26H. typicalis104
1.2   数据处理
在使用CNN训练数据集之前，对原始图像数据作了数据角度和数据增强处理，主要是为了防止过拟合和提高模型性能等。
1.2.1   数据角度处理
本文数据集图像样本由不同的作者提供，来源于不同作者的牙形刺主齿方向不尽相同，沿逆时针方向大致可以分为0°、90°、180°和270°。有研究表明，在不提前对数据进行增强的情况下，尽管CNN通过数据增强和卷积—池化操作后会表现出小角度的旋转不变性，但是在角度旋转很大的时候，CNN的旋转不变性往往会失效[15]，导致的结果就是训练所得的分类器效果不佳。因此，为了提高分类器的准确率，最好将所有牙形刺的主齿方向统一调整为同一方向。本文将所有牙形刺主齿沿逆时针方向统一调整为90°（图2a）。
1.2.2   数据增强处理
小数据集可以使用数据增强来满足CNN需要大量数据供其学习这一要求，同时也是有效降低过拟合的方式之一。在对数据集进行增强处理之前，需要将数据集划分为训练集和验证集，也可以额外划分出测试集来对模型泛化能力进行评估。数据增强只针对训练集使用。本数据集数据增强的具体操作包括颜色变换（包括亮度调整、高斯模糊）和小角度旋转（旋转角度为-15° ~ 15°）（图2b）。


图2   数据预处理示意图
Figure 2 Schematic diagram of data preprocessing
2   数据样本描述
本文提供了3个数据集，分别为由原始图像数据构成的数据集和经过数据增强后构成的数据集，所有图像数据的格式为“*.jpg”。3个数据集的构建流程如下：（1）由原始图像数据构成的数据集文件夹命名为Raw_Dataset，在Raw_Dataset下分别建立了12个子文件夹，文件夹标签分别对应Hindeodus的12个种名，将每类牙形刺图像放置对应的子文件夹中；（2）Raw_train_val_Dataset数据集文件夹之下新建名为“train”和“val”的子文件夹分别作为训练集和验证集，“train”和“val”中再分别建立了12个子文件夹，命名分别对应Hindeodus的12个种名，将Raw_Dataset中每种牙形刺的80%数据放入“train”中，剩余的20%放入“val”中；（3）Augmented_Dataset数据集是对Raw_train_val_Dataset中训练集数据进行数据增强后得到的，验证集保持不变。以上每个子文件夹中每个图像样本按照“种名_序号.jpg”进行规范化命名（图3）。


图3   部分图像数据及命名规则示意图
Figure 3 Schematic diagram of some image data and naming rules
3   数据质量控制和评估
3.1   数据质量控制
本文从数据源头上对数据质量进行了严格控制，所有数据均来自国内外权威期刊上公开发表的学术论文中的化石图版，如《古生物学报》、《古地理学报》、Palaeontology、Journal of Earth Science、Global and Planetary Change和Geobios等。此外，在初步建立好数据集后，另外请两名学生根据附录中的参考文献，对每张化石图像进行逐一检查，确保每个数据样本对应的标签准确无误。
3.2   数据质量评估
评估方法选取应用于细粒度图像分类的Bilinear-VGG16和Bilinear-ResNet18模型对Raw_Dataset数据集进行训练和验证。由于Raw_Dataset数据样本较少，如果划分为固定的训练集和验证集，不仅没有充分利用宝贵的数据进行特征学习，还会导致验证结果具有一定的随机性。因此，本次实验对Raw_Dataset没有划分训练集和验证集，而是使用分层10折（表2中K1 ~ K10）交叉验证进行训练和验证，且在训练过程中载入了模型在ImageNet数据集上训练过的权重进行迁移学习。分层10折交叉验证亦可以保证每一折训练过程中，每一类训练集样本数占该类样本总数的比例是相同的。最后，将验证结果求平均值即代表模型的性能（表2）。Bilinear-VGG16在验证集上的准确率为0.43 ~ 0.64，平均值为0.56；Bilinear-ResNet18在验证集上的准确率为0.52 ~ 0.72，平均值为0.61。考虑到实验是在少数据状态进行的，且化石数据异于常规图像数据自带一些天然缺陷（如化石并非严格完整的、部分化石表面被胶体所覆盖导致纹理结构被破坏），使用Bilinear-ResNet18在Raw_Dataset数据集上依旧取得了0.61的准确率，表明本数据是可以用于训练的。随着人工智能相关理论技术（如小样本学习）的进一步发展，相信其准确率还有较大的提升空间。
表2   Raw_Dataset数据集的实验结果
CNN验证集准确率
K1K2K3K4K5K6K7K8K9K10平均值
Bilinear-VGG160.640.630.600.620.620.440.540.430.480.620.56
Bilinear-ResNet180.670.640.680.720.700.580.520.490.540.600.61
对于Raw_train_val_Dataset数据集，使用Bilinear-ResNet18网络模型分别采取了基于参数的迁移学习和从头开始两种方式进行分别训练，结果表明使用迁移学习的模型具有更高的准确率（表3）。此外，笔者使用Bilinear-ResNet18对数据增强后的Augmented_Dataset数据集进行了训练，相比之前在Raw_train_val_Dataset数据集上获取的准确率有了一定的提升（表3），表明对牙形刺数据集进行数据增强有助于提高模型的性能。
表3   迁移学习和数据增强的实验结果
数据集是否迁移学习验证集准确率
Raw_train_val_Dataset否0.29
Raw_train_val_Dataset是0.615
Augmented_Dataset是0.663
4   数据价值
尽管现在存在一些已公开的古生物化石数据集或者线上化石展示平台，但是这些图像数据基本是粗粒度的，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图像数据集。牙形刺细粒度数据集的建立，可以为相关研究人员提供一定的数据基础。数据不仅可以直接应用于弱监督CNN模型，也可以通过专业人士添加额外的人工标注信息后，再使用强监督CNN模型进行训练。
在本数据集的基础上，相关研究人员可以继续将本次涉及的12种牙形刺数据添加到对应标签的文件夹中来增加数据，还可以增加Hindeodus属的其他种（如H. bicuspidatus, H. lobatus, H. magnus, H. priscus）。随着数据集在数量和类别上的进一步扩充，使用CNN模型和加入细粒度图像特征提取技术训练出来的分类器，可以切实应用于牙形刺Hindeodus的智能鉴定，从而有效推进化石鉴定由人工化向具有实际应用能力的智能化方向发展。
5   数据使用方法和建议
本文为用户提供了3个数据集：Raw_Dataset（原始数据且未划分训练集和验证集）、Raw_train_val_Dataset （对Raw_Dataset数据集按照8:2的比例划分训练集和验证集所得）、Augmented_Dataset（对Raw_train_val_Dataset数据集中训练集进行数据增强所得）。Raw_train_val_Dataset 和Augmented_Dataset划分了固定的训练集和验证集，用户可以采用常规训练方法进行模型训练。需要指出的是，Augmented_Datase的训练集是对Raw_Dataset中80%数据进行数据增强后得到的，如用户欲按其他比例划分数据集，可以使用Raw_Dataset自主划分训练集和验证集，然后对所划分的训练集进行数据增强操作，不需要在验证集上进行数据增强。Raw_Dataset在使用时，可以参考本文所采用的分层K折交叉验证的方法，这不仅可以充分利用本来不多且宝贵的数据，而且还能有效降低单次验证产生的随机性，从而可以更加真实的显示不同模型的性能。
针对数据集存在各类别不均衡的问题，建议在损失函数中根据各类样本数据的多寡赋予不同的权重。在训练方式上，作者发现相比从头训练，基于参数的迁移学习能明显提高模型的准确率。
附 录
牙形刺数据来源参考文献：
[1] METCALFE I, CROWLEY J L. Upper Permian and Lower Triassic conodonts, high-precision U-Pb zircon ages and the Permian-Triassic boundary in the Malay Peninsula[J]. Journal of Asian Earth Sciences, 2020, 199: 104403. DOI: 10.1016/j.jseaes.2020.104403.
[2] JIANG H S, LAI X L, LUO G M, et al. Restudy of conodont zonation and evolution across the P/T boundary at Meishan section, Changxing, Zhejiang, China[J]. Global and Planetary Change, 2007, 55(1/2/3): 39–55. DOI: 10.1016/j.gloplacha.2006.06.007.
[3] MEI S L, ZHANG K X, WARDLAW B R. A refined succession of Changhsingian and Griesbachian neogondolellid conodonts from the Meishan section, candidate of the global stratotype section and point of the Permian-Triassic boundary[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 1998, 143(4): 213–226. DOI: 10.1016/S0031-0182(98)00112-6.
[4] NICOLL R S, METCALFE I, WANG C Y. New species of the conodont Genus Hindeodus and the conodont biostratigraphy of the Permian-Triassic boundary interval[J]. Journal of Asian Earth Sciences, 2002, 20(6): 609–631. DOI: 10.1016/S1367-9120(02)00021-4.
[5] JIANG H S, LAI X L, YAN C B, et al. Revised conodont zonation and conodont evolution across the Permian-Triassic boundary at the Shangsi section, Guangyuan, Sichuan, South China[J]. Global and Planetary Change, 2011, 77(3/4): 103–115. DOI: 10.1016/j.gloplacha.2011.04.003.
[6] JI Z S, YAO J X, ISOZAKI Y, et al. Conodont biostratigraphy across the Permian–Triassic boundary at Chaotian, in northern Sichuan, China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2007(1/2), 252: 39–55. DOI: 10.1016/J.PALAEO.2006.11.033.
[7] 袁东勋, 沈树忠. 重庆中梁山凉风垭二叠-三叠系界线附近牙形类生物地层研究[J]. 古生物学报, 2011, 50(4): 420–438. DOI:10.19800/j.cnki.aps.2011.04.002. [YUAN D X, SHEN S Z. Conodont succession across the Permian-Triassic boundary of the liangfengya section, Chongqing, South China[J]. Acta Palaeontologica Sinica, 2011, 50(4): 420–438. DOI: 10.19800/j.cnki.aps.2011.04.002.]
[8] YUAN D X, CHEN J, ZHANG Y C, et al. Changhsingian conodont succession and the end-Permian mass extinction event at the Daijiagou section in Chongqing, Southwest China[J]. Journal of Asian Earth Sciences, 2015, 105: 234–251. DOI: 10.1016/j.jseaes.2015.04.002.
[9] WANG G Q, XIA W C. Conodont zonation across the Permian–Triassic boundary at the Xiakou section, Yichang City, Hubei Province and its correlation with the Global Stratotype Section and Point of the PTB[J]. Canadian Journal of Earth Sciences, 2004, 41(3): 323–330. DOI: 10.1139/E04-008.
[10] BAI R Y, DAI X, SONG H J. Conodont and ammonoid biostratigraphies around the Permian-Triassic boundary from the jianzishan of South China[J]. Journal of Earth Science, 2017, 28(4): 595–613. DOI: 10.1007/s12583-017-0754-4.
[11] WANG L N, WIGNALL P B, WANG Y B, et al. Depositional conditions and revised age of the Permo-Triassic microbialites at Gaohua section, Cili County (Hunan Province, South China)[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2016, 443: 156–166. DOI: 10.1016/j.palaeo.2015.11.032.
[12] BROSSE M, BUCHER H, BAGHERPOUR B, et al. Conodonts from the Early Triassic microbialite of Guangxi (South China): implications for the definition of the base of the Triassic System[J]. Palaeontology, 2015, 58(3): 563–584. DOI:10.1111/pala.12162.
[13] 陈军, Charles M. Henderson, 沈树忠. 浙江黄芝山剖面二叠-三叠系界线附近的牙形类序列及其地层对比(英文)[J]. 古生物学报, 2008, 47(1): 91–114. [CHEN J, HENDERSON C, SHEN S Z. Conodont succession around the Permian-Triassic boundary at the Huangzhishan section, Zhejiang and its stratigraphic correlation[J]. Acta Palaeontologica Sinica, 2008, 47(1): 91–114.]
[14] SUN D Y, TONG J N, XIONG Y L, et al. Conodont biostratigraphy and evolution across Permian-Triassic boundary at Yangou Section, Leping, Jiangxi Province, South China[J]. Journal of Earth Science, 2012, 23(3): 311–325. DOI: 10.1007/s12583-012-0255-4.
[15] YANG B, LAI X L, WIGNALL P B, et al. A newly discovered earliest Triassic chert at Gaimao section, Guizhou, southwestern China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2012, 344/345: 69–77. DOI: 10.1016/j.palaeo.2012.05.019.
[16] YAN C B, WANG L N, JIANG H, et al. Uppermost Permian to lower Triassic conodonts at Bianyang section, guihzou Province, South China[J]. Palaios, 2013, 28(7/8): 509–522. DOI:10.2110/palo.2012.p12-077r.
[17] ZHANG N, JIANG H S, ZHONG W L, et al. Conodont biostratigraphy across the Permian-Triassic boundary at the Xinmin section, Guizhou, South China[J]. Journal of Earth Science, 2014, 25(5): 779–786. DOI: 10.1007/s12583-014-0472-0.
[18] CHEN J, BEATTY T W, HENDERSON C M, et al. Conodont biostratigraphy across the Permian-Triassic boundary at the Dawen section, Great Bank of Guizhou, Guizhou Province, South China: implications for the Late Permian extinction and correlation with Meishan[J]. Journal of Asian Earth Sciences, 2009, 36(6): 442–458. DOI: 10.1016/j.jseaes.2008.08.002.
[19] CHEN Y L, JIANG H S, LAI X L, et al. Early Triassic conodonts of Jiarong, nanpanjiang basin, southern Guizhou Province, South China[J]. Journal of Asian Earth Sciences, 2015, 105: 104–121. DOI: 10.1016/j.jseaes.2015.03.014.
[20] LIANG L, TONG J N, SONG H J, et al. Lower-Middle Triassic conodont biostratigraphy of the mingtang section, nanpanjiang basin, South China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2016, 459: 381–393. DOI: 10.1016/j.palaeo.2016.07.027.
[21] METCALFE I, NICOLL R S. Conodont biostratigraphic control on transitional marine to non-marine Permian-Triassic boundary sequences in Yunnan-Guizhou, China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2007, 252(1/2): 56–65. DOI: 10.1016/j.palaeo.2006.11.034.
[22] WANG L N, WIGNALL P B, SUN Y D, et al. New Permian-Triassic conodont data from Selong (Tibet) and the youngest occurrence of Vjalovognathus[J]. Journal of Asian Earth Sciences, 2017, 146: 152–167. DOI: 10.1016/j.jseaes.2017.05.014.
[23] YUAN D X, ZHANG Y C, SHEN S Z. Conodont succession and reassessment of major events around the Permian-Triassic boundary at the Selong Xishan section, southern Tibet, China[J]. Global and Planetary Change, 2018, 161: 194–210. DOI: 10.1016/j.gloplacha.2017.12.024.
[24] WU G C, JI Z S, TROTTER J A, et al. Conodont biostratigraphy of a new Permo-Triassic boundary section at Wenbudangsang, north Tibet[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2014, 411: 188–207. DOI: 10.1016/j.palaeo.2014.06.016.
[25] PFRRI, M C, FARABEGOLI E. Conodonts across the Permian-Triassic boundary in the Southern Alps[J]. Courier Forschungsinstitut Senckenberg, 2003, 245: 281–313.
[26] KOZUR H W. Biostratigraphy and event stratigraphy in Iran around the Permian-Triassic Boundary (PTB): implications for the causes of the PTB biotic crisis[J]. Global and Planetary Change, 2007, 55(1/2/3): 155–176. DOI: 10.1016/j.gloplacha.2006.06.011.
[27] WARDLAW B R, NESTELL M K, NESTELL G P, et al. Conodont biostratigraphy of the Permian-Triassic boundary sequence at Lung Cam, Vietnam[J]. Micropaleontology, 2015, 61(4/5): 313–334. DOI: 10.47894/mpal.61.4.05.
[28] METCALFE I. Changhsingian (Late Permian) conodonts from Son La, northwest Vietnam and their stratigraphic and tectonic implications[J]. Journal of Asian Earth Sciences, 2012, 50: 141–149. DOI: 10.1016/j.jseaes.2012.01.002.
[29] KOLAR-JURKOVSEK T, JURKOVSEK B, ALJINOVIC D. Conodont biostratigraphy and lithostratigraphy across the Permian-Triassic boundary at the lukac section in western Slovenia[J]. Rivista Italiana Di Paleontologia e Stratigrafia, 2011, 117(1): 115–133.
[30] SUDAR M, PERRI M C, HAAS J. Conodonts across the Permian-Triassic boundary in the Bükk mountains (NE Hungary)[J]. Geologica Carpathica, 2008, 59(6): 491–502.
[31] SUDAR M, JOVANOVIC D, KOLAR-JURKOVŠEK T. Late Permian conodonts from jadar block (Vardar zone, northwestern Serbia)[J]. Geologica Carpathica, 2007, 58(2): 145–152.
[32] PERRI M C, MOLLOY P D, TALENT J A. Earliest Triassic conodonts from Chitral, northernmost Pakistan[J]. Rivista Italiana Di Paleontologia e Stratigrafia, 2004, 110(2): 467–478. DOI: 10.13130/2039-4942/5817.
[33] ALGEO T, HENDERSON C M, ELLWOOD B, et al. Evidence for a diachronous Late Permian marine crisis from the Canadian Arctic region[J]. Geological Society of America Bulletin, 2012, 124(9–10): 1424–1448. DOI:10.1130/b30505.1.
[34] LYU Z Y, ORCHARD M J, CHEN Z Q, et al. Uppermost Permian to lower Triassic conodont successions from the Enshi area, western Hubei Province, South China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2019, 519: 49–64. DOI: 10.1016/j.palaeo.2017.08.015.
[35] ZHANG Y, ZHANG K X, SHI G R, et al. Restudy of conodont biostratigraphy of the Permian-Triassic boundary section in Zhongzhai, southwestern Guizhou Province, South China[J]. Journal of Asian Earth Sciences, 2014, 80: 75–83. DOI: 10.1016/j.jseaes.2013.10.032.
[36] YUAN D X, SHEN S Z, HENDERSON C M, et al. Revised conodont-based integrated high-resolution timescale for the Changhsingian Stage and end-Permian extinction interval at the Meishan sections, South China[J]. Lithos, 2014, 204: 220–245. DOI: 10.1016/j.lithos.2014.03.026.
[37] METCALFE I, NICOLL R S, WARDLAW B R. Conodont index fossil Hindeodus changxingensis Wang fingers greatest mass extinction event[J]. Palaeoworld, 2007, 16(1/2/3): 202–207. DOI:10.1016/j.palwor.2007.01.001.
[38] JIANG H S, LAI X L, SUN Y D, et al. Permian-Triassic conodonts from Dajiang (Guizhou, South China) and their implication for the age of microbialite deposition in the aftermath of the End-Permian mass extinction[J]. Journal of Earth Science, 2014, 25(3): 413–430. DOI: 10.1007/s12583-014-0444-4.
[39] GHADERI A, LEDA L, SCHOBBEN M, et al. High-resolution stratigraphy of the changhsingian (late Permian) successions of NW Iran and the transcaucasus based on lithological features, conodonts and ammonoids[J]. Fossil Record, 2014, 17(1): 41–57. DOI: 10.5194/fr-17-41-2014.
[40] ZHANG L, ORCHARD M J, ALGEO T J, et al. An intercalibrated Triassic conodont succession and carbonate carbon isotope profile, Kamura, Japan[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2019, 519: 65–83. DOI: 10.1016/j.palaeo.2017.09.001.
[41] BROSSE M, BAUD A, BHAT G M, et al. Conodont-based Griesbachian biochronology of the Guryul Ravine section (basal Triassic, Kashmir, India)[J]. Geobios, 2017, 50(5/6): 359–387. DOI: 10.1016/j.geobios.2017.10.001.
[42] LEHRMANN D J, STEPCHINSKI L, ALTINER D, et al. An integrated biostratigraphy (conodonts and foraminifers) and chronostratigraphy (paleomagnetic reversals, magnetic susceptibility, elemental chemistry, carbon isotopes and geochronology) for the Permian-Upper Triassic strata of Guandao section, Nanpanjiang Basin, South China[J]. Journal of Asian Earth Sciences, 2015, 108: 117–135. DOI: 10.1016/j.jseaes.2015.04.030.
[43] ZHANG K X, TONG J N, SHI G R, et al. Early Triassic conodont-palynological biostratigraphy of the Meishan D section in Changxing, Zhejiang Province, South China[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2007, 252(1/2): 4–23. DOI: 10.1016/j.palaeo.2006.11.031.
[44] KOLAR-JURKOVŠEK T, JURKOVŠEK B, NESTELL G P, et al. Biostratigraphy and sedimentology of upper Permian and lower Triassic strata at masore, western Slovenia[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2018, 490: 38–54. DOI: 10.1016/j.palaeo.2017.09.013.
[45] KOLAR-JURKOVŠEK T, JURKOVŠEK B. First record of Hindeodus - Isarcicella population in Lower Triassic of Slovenia[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2007, 252(1/2): 72–81. DOI: 10.1016/j.palaeo.2006.11.036.
[46] YANG B, LI H X, WIGNALL P B, et al. Latest wuchiapingian to earliest Triassic conodont zones and δ13Ccarb isotope excursions from deep-water sections in western Hubei Province, South China[J]. Journal of Earth Science, 2019, 30(5): 1059–1074. DOI: 10.1007/s12583-019-1018-2.
[47] KERSHAW S, GUO L, SWIFT A, et al. ?Microbialites in the Permian-Triassic boundary interval in central China: structure, age and distribution[J]. Facies, 2002, 47(1): 83–89. DOI: 10.1007/BF02667707.
[48] TANG H, KERSHAW S, LIU H, et al. Permian-Triassic boundary microbialites (PTBMs) in southwest China: implications for paleoenvironment reconstruction[J]. Facies, 2017, 63(1): 2. DOI: 10.1007/s10347-016-0482-8.
[49] GAETANI M, NICORA A, HENDERSON C, et al. Refinements in the upper Permian to lower Jurassic stratigraphy of Karakorum, Pakistan[J]. Facies, 2013, 59(4): 915–948. DOI: 10.1007/s10347-012-0346-9.
[50] KOLAR-JURKOVŠEK T, HRVATOVIĆ H, ALJINOVIĆ D, et al. Permian-Triassic biofacies of the teočak section, Bosnia and Herzegovina[J]. Global and Planetary Change, 2021, 200: 103458. DOI: 10.1016/j.gloplacha.2021.103458.
[51] MU X N, KERSHAW S, LI Y, et al. High-resolution carbon isotope changes in the Permian-Triassic boundary interval, Chongqing, South China; implications for control and growth of earliest Triassic microbialites[J]. Journal of Asian Earth Sciences, 2009, 36(6): 434–441. DOI: 10.1016/j.jseaes.2007.08.004.
[52] RICHOZ S, KRYSTYN L, BAUD A, et al. Permian-Triassic boundary interval in the Middle East (Iran and N. Oman): progressive environmental change from detailed carbonate carbon isotope marine curve and sedimentary evolution[J]. Journal of Asian Earth Sciences, 2010, 39(4): 236–253. DOI: 10.1016/j.jseaes.2009.12.014.
[53] YANG B, YUAN D X, HENDERSON C M, et al. Parafurnishius, an Induan (Lower Triassic) conodont new genus from northeastern Sichuan Province, southwest China and its evolutionary implications[J]. Palaeoworld, 2014, 23(3/4): 263–275. DOI: 10.1016/j.palwor.2014.10.003.
[54] YUAN D X, SHEN S Z, HENDERSON C M, et al. Integrative timescale for the Lopingian (Late Permian): a review and update from Shangsi, South China[J]. Earth-Science Reviews, 2019, 188: 190–209. DOI: 10.1016/j.earscirev.2018.11.002.
[55] XIAO Y F, WU K, TIAN L, et al. Framboidal pyrite evidence for persistent low oxygen levels in shallow-marine facies of the Nanpanjiang Basin during the Permian-Triassic transition[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2018, 511: 243–255. DOI: 10.1016/j.palaeo.2018.08.012.
[56] MAALEKI-MOGHADAM M, RAFIEI B, RICHOZ S, et al. Anachronistic facies and carbon isotopes during the end-Permian biocrisis: evidence from the mid-Tethys (Kisejin, Iran)[J]. Palaeogeography, Palaeoclimatology, Palaeoecology, 2019, 516: 364–383. DOI: 10.1016/j.palaeo.2018.12.007.
[57] KOZUR H W. The conodonts Hindeodus isarcicella and sweetohindeodus in the uppermost Permian and lowermost Triassic[J]. Geologia Croatica, 2010, 49(1): 81–115.
[58] 陈军, Charles M.Henderson, 沈树忠. 浙江黄芝山剖面二叠-三叠系界线附近的牙形类序列及其地层对比[J]. 古生物学报, 2008, 47(1): 91–114. DOI: 10.3969/j.issn.0001-6616.2008.01.007. [CHEN J, HENDERSON C, SHEN S Z. Conodont succession around the Permian-Triassic boundary at the Huangzhishan section, Zhejiang and its stratigraphic correlation[J]. Acta Palaeontologica Sinica, 2008, 47(1): 91–114. DOI: 10.3969/j.issn.0001-6616.2008.01.007.]
[59] 袁东勋, 沈树忠. 重庆中梁山凉风垭二叠-三叠系界线附近牙形类生物地层研究[J]. 古生物学报, 2011, 50(4): 420–438. DOI: 10.19800/j.cnki.aps.2011.04.002. [YUAN D X, SHEN S Z. Conodont succession across the Permian-Triassic boundary of the liangfengya section, Chongqing, South China[J]. Acta Palaeontologica Sinica, 2011, 50(4): 420–438. DOI: 10.19800/j.cnki.aps.2011.04.002.]
[60] 刘建波, 江崎洋一, 杨守仁, 等. 贵州罗甸二叠纪末生物大灭绝事件后沉积的微生物岩的时代和沉积学特征[J]. 古地理学报, 2007, 9(5): 473–486. [LIU J B, EZAKI Y, YANG S R, et al. Age and sedimentology of microbialites after the end-Permian mass extinction in Luodian, Guizhou Province[J]. Journal of Palaeogeography, 2007, 9(5): 473–486.]
致 谢
感谢西华师范大学地理科学学院鲜萍和温欣宜两位同学对数据标签的细致检查。
[1]
杜远生, 童金南. 古生物地史学概论[M]. 2版. 武汉: 中国地质大学出版社, 2009. [DU Y S, TONG J N. Introduction to palaeontology and historical geology[M]. 2nd ed. Wuhan: China University of Geosciences Press, 2009.]
+ CSCD · Baidu Scholar
[2]
王成源, 王志浩. 中国牙形刺生物地层[M]. 杭州: 浙江大学出版社, 2016. [WANG C Y, WANG Z H. Conodont biostratigraphy in China[M]. Hangzhou: Zhejiang University Press, 2016.]
+ CSCD · Baidu Scholar
[3]
周永章, 张良均, 张奥多. 地球科学大数据挖掘与机器学习[M]. 广州: 中山大学出版社, 2018. [ZHOU Y Z, ZHANG L J, ZHANG A D. Big data mining & machine learning in geoscience[M]. Guangzhou: Sun Yat-sen University Press, 2018.]
+ CSCD · Baidu Scholar
[4]
LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521(7553): 436–444. DOI:10.1038/nature14539.
+ CSCD · Baidu Scholar
[5]
BOUREL B, MARCHANT R, DE GARIDEL-THORON T, et al. Automated recognition by multiple convolutional neural networks of modern, fossil, intact and damaged pollen grains[J]. Computers & Geosciences, 2020, 140: 104498. DOI:10.1016/j.cageo.2020.104498.
+ CSCD · Baidu Scholar
[6]
HSIANG A Y, BROMBACHER A, RILLO M C, et al. Endless forams: >34, 000 modern planktonic foraminiferal images for taxonomic training and automated species recognition using convolutional neural networks[J]. Paleoceanography and Paleoclimatology, 2019, 34(7): 1157–1177. DOI: 10.1029/2019PA003612.
+ CSCD · Baidu Scholar
[7]
MITRA R, MARCHITTO T M, GE Q, et al. Automated species-level identification of planktic foraminifera using convolutional neural networks, with comparison to human performance[J]. Marine Micropaleontology, 2019, 147: 16–24. DOI:10.1016/j.marmicro.2019.01.005.
+ CSCD · Baidu Scholar
[8]
KEÇELI A S, KEÇELI S U, KAYA A. Classification of radiolarian fossil images with deep learning methods[C]//2018 26th Signal Processing and Communications Applications Conference (SIU). Izmir, Turkey. IEEE, 2018: 1–4. DOI:10.1109/SIU.2018.8404460.
+ CSCD · Baidu Scholar
[9]
WANG H Z, LI C F, ZHANG Z F, et al. Fossil brachiopod identification using a new deep convolutional neural network[J]. Gondwana Research, 2022, 105: 290–298. DOI:10.1016/j.gr.2021.09.011.
+ CSCD · Baidu Scholar
[10]
安玉钏, 陈雁, 黄玉楠, 等. 基于深度学习的介形类化石层次化识别[J]. 地质论评, 2022, 68(2): 673–684. DOI: 10.16509/j.georeview.2021.11.031. [AN Y C, CHEN Y, HUANG Y N, et al. Hierarchical recognition of ostracod fossils based on deep learning[J]. Geological Review, 2022, 68(2): 673–684. DOI: 10.16509/j.georeview.2021.11.031.]
+ CSCD · Baidu Scholar
[11]
柳天滋, 陈昕, 李想, 等. 基于深度残差神经网络迁移学习的牙形刺图像识别[J]. 古生物学报, 2020, 59(4): 512–523. DOI: 10.19800/j.cnki.aps.2020.042. [LIU T Z, CHEN X, LI X, et al. Conodont image recognition based on transfer learning of deep residual neural network[J]. Acta Palaeontologica Sinica, 2020, 59(4): 512–523. DOI: 10.19800/j.cnki.aps.2020.042.]
+ CSCD · Baidu Scholar
[12]
徐卉清, 樊隽轩, 杨娇, 等. 应用卷积神经网络进行化石图像分类[C]//中国古生物学会第十二次全国会员代表大会暨第29届学术年会论文摘要集. 郑州, 2018: 178–179. [XU H Q, FAN J X, YANG J, et al. Classification of fossil images using convolutional neural networks[C]//Abstract Volume, Joint Meetings on the 12th National Congress of the Palaeontological Society of China (PSC) and the 29th Annual Conference of PSC. Zhengzhou, 2018: 178–179.]
+ CSCD · Baidu Scholar
[13]
LIN T Y, ROYCHOWDHURY A, MAJI S. Bilinear CNN models for fine-grained visual recognition[C]//2015 IEEE International Conference on Computer Vision (ICCV). Santiago, Chile. IEEE, 2016: 1449–1457. DOI:10.1109/ICCV.2015.170.
+ CSCD · Baidu Scholar
[14]
FU J L, ZHENG H L, MEI T. Look closer to see better: recurrent attention convolutional neural network for fine-grained image recognition[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu, HI, USA. IEEE, 2017: 4476–4484. DOI:10.1109/CVPR.2017.476.
+ CSCD · Baidu Scholar
[15]
ZHOU Y Z, YE Q X, QIU Q, et al. Oriented response networks[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA. IEEE, 2017: 4961–4970. DOI:10.1109/CVPR.2017.527.
+ CSCD · Baidu Scholar
数据引用格式
段雄. Hindeodus牙形刺数据集[DS/OL]. Science Data Bank, 2022. (2022-12-22). DOI: 10.57760/sciencedb.j00001.00641.
Baidu
稿件与作者信息
论文引用格式
段雄. 应用卷积神经网络分类的Hindeodus牙形刺细粒度数据集[J/OL]. 中国科学数据, 2023, 8(2). (2023-04-06). DOI: 10.11922/11-6035.csd.2022.0075.zh.
段雄
DUAN Xiong
数据集的构建、技术支持及论文撰写。
duanxiong00@163.com
（1988—），男，湖北洪湖人，博士，讲师，研究方向为计算机地球科学相关研究。
四川省自然科学基金（2022NSFSC1177）；西华师范大学博士科研启动项目（20E031）
Natural Science Foundation of Sichuan Province (2022NSFSC1177); Doctoral Research Project of China West Normal University (20E031).
Baidu
出版历史
I区发布时间：2022年12月2日 （ 版本ZH1）
II区出版时间：2023年4月6日 （ 版本ZH2）
最近更新时间：2023年4月6日 （ 版本ZH3）
CSCD  ·  Badiu Scholar
参考文献列表中查看
中国科学数据
csdata

  • --

中国科学院期刊云服务平台 · 中国科学院科学传播局
技术支持 中国科学院计算机网络信息中心 · 印刻学术
Baidu
map