【摘要】 在手掌静脉图像采集的过程中易受手掌摆放姿势、光源条件等外界因素的影响,造成识别准确度欠佳。为了提高手掌静脉图像识别的精准度和鲁棒性,提出一种基于改进AlexNet深度卷积神经网络的手掌静脉识别方法。该方法首先通过图像分割、指根关键点定位、感兴趣区域图像提取等三个阶段对采集的手掌静脉图像进行预处理;其次,针对人体手掌静脉识别的应用场景,通过适当调整经典的AlexNet卷积神经网络的结构并对卷积层的输出进行批标准化操作,同时,将深度学习理论中的注意力机制应用到该网络中,进而优化AlexNet神经网络,使用优化后的AlexNet神经网络对预处理后的图像自动进行特征提取、分类和识别;最后,在公开的Polyu和CASIA多光谱掌纹数据集上进行大量的实验,达到的最佳识别率分别为99.93%和99.51%,实验验证了所提方法的有效性。更多还原

【Abstract】 In the process of palm vein image acquisition,palm vein images are susceptible to external factors like palm posture and light source conditions,which will result in poor recognition accuracy. Therefore,a palm vein recognition method based on improved AlexNet depth convolution neural network is proposed to improve the accuracy and robustness of the image recognition. Firstly,the collected palm vein image is preprocessed by image segmentation,finger root key point location and image extraction in the region of interest(ROI). Secondly,according to the application context of palm vein recognition,the classical AlexNet convolution neural network structure is adjusted appropriately, and the output of convolution layer is standardized in batches. The attention mechanism in deep learning theory is applied to the above-mentioned network to optimize the AlexNet neural network. The optimized AlexNet neural network is used to automatically extract,classify and identify the features of the preprocessed images. A large number of experiments were performed on public PolyU and CASIA multispectral palmprint datasets. The optimal recognition rates are 99.93% and 99.51% respectively. The experiments verify the effectiveness of this method.更多还原