Odonata identification using Customized Convolutional Neural Networks

作者：

Highlights：

• We introduce a large-scale research-grade odonates image dataset.

• Detailed investigation on CNN models for various input image resolutions.

• Customized DenseNet161 (450px × 450px) had the highest top-1 accuracy.

• Class Activation Map analysis was performed to evaluate the CNN models.

摘要

•We introduce a large-scale research-grade odonates image dataset.•Detailed investigation on CNN models for various input image resolutions.•Customized DenseNet161 (450px × 450px) had the highest top-1 accuracy.•Class Activation Map analysis was performed to evaluate the CNN models.

论文关键词：Odonatanet,Dragonflies,Damselflies,Deep learning,Convolutional Neural Networks

论文评审过程：Received 17 February 2022, Revised 9 May 2022, Accepted 28 May 2022, Available online 4 June 2022, Version of Record 18 June 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117688