Evolutionary deep learning based on deep convolutional neural network for anime storyboard recognition
Evolutionary deep learning based on deep convolutional neural network for anime storyboard recognition
复制标题
DOI：
10.1016/j.neucom.2018.05.124
复制
发表时间：
2019-04
期刊：
影响因子：
6
通讯作者：
Saya Fujino;Taichi Hatanaka;N. Mori;Keinosuke Matsumoto
中科院分区：
文献类型：
--
作者：
Saya Fujino;Taichi Hatanaka;N. Mori;Keinosuke Matsumoto
最近，基于深度学习的图像识别得到了相当大的研究关注。在这项研究中，我们专注于动漫故事板，并将深度卷积神经网络（DCNN）应用于这些数据。然而，通过网格搜索方法来调整DCNN超参数是困难的。为此，提出了一种基于遗传算法的进化深度学习（evoDL）方法，并以真实的动漫故事板识别问题为例，通过计算机仿真验证了evoDL的有效性.
Recently, image recognition based on deep learning has gained considerable research attention.In this study, we focus on anime storyboards and apply deep convolutional neural networks (DCNNs) to those data. However, it is difficult to tune the DCNN hyperparameters by the gird search method. Therefore, we propose a novel method called evolutionary deep learning (evoDL) by adopting a genetic algorithm (GA) to solve this problem.The effectiveness of evoDL is validated by computer simulations by taking a real anime storyboard recognition problem as an example.