Deep dictionary learning with reconstruction for texture recognition

基于深度字典的纹理识别重建

阅读:1

Abstract

Texture recognition underpins critical applications in industrial quality control, robotic manipulation, and biomedical imaging. Traditional deep dictionary learning methods for texture recognition often emphasize deep feature extraction. However, they tend to lose crucial features as model depth increases, which can reduce their overall effectiveness. To address this issue, we propose a dictionary-reconstruction-based deep learning approach by incorporating a novel hybrid fusion method designed to enhance the accuracy of texture recognition. Our approach involves the successive fusion of multimodality and multi-level features. By reconstructing dictionaries learned at different levels, we integrate both deep and intuitive features. Additionally, we introduce a grouping optimization technique, based on single-sample learning, to train these reconstructed dictionaries, thereby improving feature learning and training efficiency. The proposed approach fuses feature data from various multimodal sources and constructs dictionaries at different learning levels, which enables effective feature fusion across these levels. We evaluate our approach against recent deep learning methods by using the LMT-108 and SpectroVision datasets. The results demonstrate its 97.7% and 89.4% accuracy rates, respectively, outperforming its peers and validating its robustness when handling diverse and challenging data.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。