摘要
目前,高分辨率掌纹图像识别存在细节特征点匹配法算法复杂,人工提取特征困难等问题。由于卷积神经网路的标量神经元无法获得特征之间的位置关系,应用于高分辨率掌纹图像识别效果并不理想。本文提出了一种基于改进的胶囊网络的高分辨率掌纹图像识别算法,通过去掉重构网络来换取模型体量的精简和运算速度的提升,在有限的精度损失下大大降低了算法复杂度。同时采用超深度小卷积神经网路来优化特征提取部分,为路由算法提供更优质的胶囊。由于路由算法对掌纹特征的方位比较敏感,在主胶囊层前面加入通道注意力机制以增加重要特征的权重,进一步提高识别能力。实验证明,本文改进后的胶囊网络对高分辨率掌纹图像的识别准确率可达到88.13%,识别精度和运算速度均优于基础胶囊网络方法。
At present, there are some problems in the recognition of high resolution palmprint images, such as complex algorithm of detail feature point matching and difficulty in manually extracting features. Since the scalar neurons of the convolutional neural network cannot obtain the positional relationship between the features, the recognition effect of high-resolution palmprint images is not ideal. This paper proposes a high-resolution palmprint image recognition algorithm based on an improved capsule network. By removing the reconstruction network in exchange for the simplification of the model volume and the improvement of the calculation speed, the complexity of the algorithm was greatly reduced under the limited accuracy loss.At the same time, the ultra-deep small convolutional neural network was used to optimize the feature extraction part to provide better capsules for the routing algorithm. Since the routing algorithm is more sensitive to the location of palmprint features, a channel attention mechanism was added in front of the main capsule layer to increase the weight of important features and further improve the recognition ability. Experiments show that the recognition accuracy of the improved capsule network for high-resolution palmprint images can reach 88.13%,and the recognition accuracy and operation speed are better than the basic capsule network method..
作者
吴碧巧
王天一
WU Bi-qiao;WANG Tian-yi(College of Big Data and Information Engineering,Guizhou University,Guiyang Guizhou 550025,China)
出处
《计算机仿真》
北大核心
2022年第9期234-238,共5页
Computer Simulation
基金
黔科合平台人才([2018]5616)。
关键词
高分辨率掌纹图像
胶囊网络
卷积网络
动态路由算法
注意力
High-resolution palmprint images
Capsule network
Convolution network
Dynamic routing algorithm
Attention
作者简介
吴碧巧(1994-),女(汉族),重庆市永川区人,硕士研究生,主要研究领域为图像识别;王天一(1989-),男(汉族),辽宁省锦州市人,副教授,硕士研究生导师,主要研究领域为量子通信,图像处理,计算机视觉。
