基于分组LSTM与CNN的青铜器锈蚀类别智能标识方法

Intelligent identification method of bronze rust category based on grouping LSTM and CNN

青铜器锈蚀标识在青铜器保护修复中具有重要作用。目前,青铜器锈蚀标识以人工标识为主,但是,由于该青铜器锈蚀较为混杂,不易识别,也会因个人经验差异出现标注不一致等问题。因此,该文提出了一种青铜器锈蚀智能标识方法,即基于分组LSTM与CNN的标识方法(multi-group LSTM and CNN,MGLC)。该方法对青铜器高光谱图像设计了一种端到端的网络,该网络在CNN网络的基础上引入了多分组策略的LSTM网络,CNN网络可以提取丰富的空间信息,多分组策略的LSTM可以探索光谱的上下文信息,MGLC网络有效地结合了青铜器高光谱图像的空间信息和光谱信息。实验表明,该方法在青铜器锈蚀类别的智能标识中具有良好的效果。

The rust mark of bronze ware plays an important role in the protection and repair of bronze ware.At present,the bronze rust identification is mainly manual identification,but due to the mixed bronze rust,it is difficult to identify,and there are problems such as inconsistent identification due to personal experience.Therefore,this paper proposes an intelligent identification method for bronze rust,that is,the identification method based on grouping LSTM and CNN(multi-group LSTM and CNN,MGLC).This method is an end-to-end network for bronze hyperspectral image design.Based on CNN network,the network introduces multi grouping strategy LSTM network.CNN network can extract rich spatial information,multi grouping strategy LSTM can explore spectral context information,and MGLC network effectively combines spatial information and spectral information of bronze hyperspectral image.Experiments show that this method has a good effect in the intelligent identification of bronze rust category.