Deep convolutional networks for human sketches by means of the evolutionary deep learning

Recognition of human sketches is one of the most interesting and difficult issues in image recognition. Recently, deep convolutional neural networks (DCNNs) have been successfully applied to various image recognition tasks. Though the DCNN is a very powerful method, the high computational effort required to tune its hyperparameters represents a critical problem. In this paper, we propose a novel method called evolutionary deep learning (evoDL) that uses a genetic algorithm in order to obtain effective deep learning networks. The generalization ability of the network structure obtained using the proposed method is confirmed by a computer experiment.