Automated identification of benthic epifauna with computer vision
Automated identification of benthic epifauna with computer vision
复制标题
DOI：
10.3354/meps12925
复制
发表时间：
2019-04
影响因子：
2.5
通讯作者：
Nils Piechaud;Christopher Hunt;P. Culverhouse;N. Foster;K. Howell
中科院分区：
文献类型：
--
作者：
Nils Piechaud;Christopher Hunt;P. Culverhouse;N. Foster;K. Howell
：底栖生态系统长期采样不足，特别是在13个大于50米的环境中。然而，人为威胁水平的上升使得数据收集变得更加紧迫。目前，现代水下采样工具，15特别是自主水下航行器（AUV），能够收集大量图像16数据集，但无法绕过手动图像注释形成的瓶颈。17计算机视觉（CV）提供了一种更快、更一致、更具成本效益和可共享的18替代手动注释的方法。我们使用Tensorflow来评估Inception V3模型在不同数量的训练图像下的性能，并评估它可以区分多少不同的类（分类）。分类器（模型）21用增加的数据量（每个分类群的20至1000个图像）训练，22
: Benthic ecosystems are chronically undersampled, particularly in 13 environments >50m. Yet, a rising level of anthropogenic threats makes data 14 collection ever more urgent. Currently, modern underwater sampling tools, 15 particularly Autonomous Underwater Vehicles (AUV), are able to collect vast image 16 datasets, but cannot bypass the bottleneck formed by manual image annotation. 17 Computer Vision (CV) offers a faster, more consistent, cost effective and a sharable 18 alternative to manual annotation. We used Tensorflow to evaluate the performance 19 of the Inception V3 model with different numbers of training images, as well as 20 assessing how many different classes (taxa) it could distinguish. Classifiers (models) 21 were trained with increasing amounts of data (20 to 1000 images of each taxa) and 22