[1]郭 锐,熊风光*,谢剑斌,等.基于改进残差池化层的纹理识别[J].计算机技术与发展,2023,33(09):37-44.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 006]
 GUO Rui,XIONG Feng-guang*,XIE Jian-bin,et al.Texture Recognition Algorithm Based on Improved Deep Residual Pooling Layer[J].,2023,33(09):37-44.[doi:10. 3969 / j. issn. 1673-629X. 2023. 09. 006]
点击复制

基于改进残差池化层的纹理识别()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
33
期数:
2023年09期
页码:
37-44
栏目:
媒体计算
出版日期:
2023-09-10

文章信息/Info

Title:
Texture Recognition Algorithm Based on Improved Deep Residual Pooling Layer
文章编号:
1673-629X(2023)09-0037-08
作者:
郭 锐12 熊风光12* 谢剑斌12 尹宇慧12 刘 磊12
1. 中北大学 大数据学院,山西 太原 030051;
2. 山西省视觉信息处理及智能机器人工程研究中心,山西 太原 030051
Author(s):
GUO Rui12 XIONG Feng-guang12* XIE Jian-bin12 YIN Yu-hui12 LIU Lei12
1. School of Data Science and Technology,North University of China,Taiyuan 030051,China;
2. Shanxi Vision Information Processing and Intelligent Robot Engineering Research Center,Taiyuan 030051,China
关键词:
纹理识别残差池化层全局最大池化多维特征融合模块多尺度特征
Keywords:
texture recognition residual pooling layer global maximum pooling multi - dimensional feature fusion module multi-scale feature
分类号:
TP391. 41
DOI:
10. 3969 / j. issn. 1673-629X. 2023. 09. 006
摘要:
纹理一直是物体图像最重要的特征之一。 针对现有纹理识别模型在复杂数据集下识别准确率不高的问题,提出一种基于改进残差池化层的纹理识别算法。 首先,提出多维特征融合模块,在纹理识别模型中同时利用高层特征和低层特征来提取更加有效的纹理特征;其次,对残差池化层进行改进,在原残差池化层的基础上,引入全局最大池化支路,为纹理识别模型增加全局空间结构观察,将原残差池化层与全局最大池化支路得到的特征向量进行拼接后作为纹理特征,提升纹理识别的准确率;再次,应用局部二值模式辅助识别策略,使用局部二值模式编码映射图像为纹理识别模型提供辅助信息;最后,将得到的纹理特征输入到分类层中,得到纹理识别结果。与现有的纹理识别方法 B-CNN、Deep filter banks、Deep TEN、TEX-Net-LF、locality-aware coding、DRP-Net 相比,该方法具有更好的纹理识别效果。
Abstract:
Texture is always one of the most important features of object images. Aiming at the low recognition accuracy of existingtexture recognition models in complex datasets,we propose a texture recognition algorithm based on improved residual pooling layer.Firstly,a multi-dimensional feature fusion module is proposed to?
extract more effective texture features by using both high-level featuresand low-level features in this texture recognition model. Secondly,the residual pooling layer is improved. On the basis of the original residual pooling layer,the global maximum pooling branch is introduced to raise the global spatial structure observation for the texture recognition model. The feature vectors obtained from the original residual pooling layer and the global maximum pooling branch are splicedas texture features to improve the accuracy of texture recognition. Thirdly,with local binary patterns aided recognition strategy,localbinary patterns encoded mapping images are used to provide auxiliary information for the texture recognition model. Finally,the obtainedtexture features are input into the classification layer to obtain the texture recognition results. The proposed method has better texture recognition effect than that of the existing texture recognition methods B-CNN,Deep filter banks,Deep TEN,TEX-Net-LF,locality-awarecoding,DRP-Net.
更新日期/Last Update: 2023-09-10