项目名称: 基于极限学习单元的多生物特征图像深度学习建模与识别研究
项目编号: No.61502338
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 杨巨成
作者单位: 天津科技大学
项目金额: 20万元
中文摘要: 深度学习是模拟人类视觉感知神经系统的认知学习,层次化、渐进式的深度学习能够获得更具表征力的高层特征,适用于多生物特征识别领域。针对传统深度学习模型存在运算效率不高的弊端,在前期研究基础上,本项目拟开展基于极限学习单元的多生物特征图像深度学习建模及识别研究。研究内容包括:①多生物特征图像的预处理以及图像数据的融合;②将极限学习融入深度学习,构建深度学习单元,并引入贝叶斯变分推理,自适应模糊粒子群优化,稀疏结构等优化算法,提高学习效率;③基于构建的极限学习单元,依次确定每层学习单元的网络参数,进行深度网络的层次化、渐进式学习;此外,针对不同的输入环境,可以融入特定的学习算法。④利用学习后的高层表征特征进行多生物特征识别。通过以上内容的研究,构建新的有效的深度学习模型,期望为多生物特征识别提供新的研究思路和方法,也为进一步深入研究打下基础。
中文关键词: 生物特征识别;深度学习;多模态生物特征识别
英文摘要: Deep learning is a kind of cognitive learning to simulate the human visual perception of the nervous system, hierarchical and incremental deep learning can get more high-level features, and it is suitable to the field of multimodal biometrics recognition. According to the low efficiency drawbacks of the traditional deep learning models, based on the preliminary studies, the project intends to do the researches on building a deep learning model and recognition for multimodal biometrics image with extreme learning units. The studies include : ① The image pre-processing and data fusion of the multimodl biometrics images; ② Fusing the extreme learning into deep learning to build the deep learning units , and combining the variational Bayesian inference, Adaptive Fuzzy Particle Swarm Optimization (AF-PSO), sparse structure optimization algorithms to improve the learning efficiency; ③Based on the built extreme learning machine units, we determine the network parameters of the learning unit of each layer to performe the hierarchical, incremental learning on the deep learning networks;Besides, it is able to fuse the certain learning methods according to the different input environments. ④Using the learned high-level representation features for multimodal biometrics recognition. By studying the above contents, we will build a new and effective deep learning model, and hope to provide new ideas and methods for the multimodal biometrics authentication, and also lay the foundation for further study.
英文关键词: biometrics;deep learning;multimodal biometrics