煤矿井下人脸识别中的特征提取与分类算法研究

发布时间：2018-05-27 13:18

本文选题：人脸识别 + 特征提取　；参考：《太原科技大学》2016年硕士论文

【摘要】：煤矿视频监控系统克服了打卡考勤中人卡不一致的缺陷,在矿工身份认证及灾后应急救援等实际应用中发挥重要作用。现有的人脸识别方法在可控环境中表现良好,但在存在遮挡变化、光照不充足等因素的非可控环境下识别效果并不可观。本文针对井下存在煤尘、粉尘、声音等各种噪声的干扰,监控识别难度加大的人脸识别的关键技术及过程展开研究,具有较大的应用价值。针对井下人脸图像易受煤尘干扰且一般局部化方法对噪声较敏感的问题,本文提出基于Shearlet变换的煤矿井下图像差异性特征提取方法。该方法利用Shearlet变换具有的方向选择特性及图像特征表示能力,并运用Shannon熵理论对特征子图赋予不同权值以及实部特征进行编码融合,有效降低了维数。实验结果表明,本文方法对受煤尘污染与噪声影响的人脸图像特征提取与抗噪能力效果良好。针对原稀疏描述方法因计算复杂度高以及对训练样本个数敏感的问题,本文提出一种差异性Shearlet特征的快速稀疏描述人脸识别方法。首先,本文提出的匹配得分融合方法在稀疏描述方法优势的基础上再次考虑了“当前”测试样本,具有“稀疏优势”;其次,根据匹配得分选取训练样本集的一个子集来表示测试样本,排除了与测试样本很不相似的训练样本对分类结果的干扰;最后,运用快速稀疏描述方法较好的解决了L0范数的优化问题。实验结果表明,本文提出的人脸识别算法在保证高识别率的同时,降低了计算复杂度。以煤矿视频监控系统为背景,本文重点研究了人脸识别流程中的图像特征提取与分类识别过程,取得一些研究进展,下一步将对Shearlet变换中分解尺度方向数与识别结果之间的关系做进一步研究,还计划将人脸识别技术与精确人员定位技术进行有效结合,更好地提高灾后救援的有效性。
[Abstract]:The video monitoring system of coal mine overcomes the defect of inconsistency between the people and the card in the process of checking on the work and attendance, and plays an important role in the practical application of mine identity authentication and emergency rescue after the disaster. The existing face recognition methods perform well in controllable environment, but the recognition effect is not significant in the uncontrollable environment with the change of occlusion and insufficient illumination. In this paper, the key technology and process of face recognition, which is difficult to monitor and recognize, is studied in view of the interference of coal dust, sound and other noises, which has great application value. Aiming at the problem that the downhole face image is susceptible to coal dust interference and the general localization method is sensitive to noise, this paper presents a method of extracting the difference feature of underground coal mine image based on Shearlet transform. This method utilizes the direction selection characteristic of Shearlet transform and the ability of image feature representation, and uses Shannon entropy theory to encode and fuse different weights and real part features of feature subgraphs, thus effectively reducing the dimension. The experimental results show that the proposed method has a good effect on feature extraction and anti-noise ability of face images affected by coal dust pollution and noise. Aiming at the problem of high computational complexity and sensitivity to the number of training samples in the original sparse description method, this paper proposes a fast sparse face recognition method based on different Shearlet features. Firstly, based on the advantages of sparse description method, the matching score fusion method proposed in this paper again considers the "current" test sample, which has "sparse advantage"; secondly, According to the matching score, a subset of the training sample set is selected to represent the test sample, which eliminates the interference of the training sample which is not similar to the test sample to the classification results. The problem of L _ 0 norm optimization is well solved by using the fast sparse description method. Experimental results show that the proposed face recognition algorithm not only guarantees high recognition rate, but also reduces the computational complexity. Based on the coal mine video surveillance system, this paper focuses on the process of image feature extraction and classification recognition in the process of face recognition. The next step will be to further study the relationship between the decomposition scale direction number and the recognition results in Shearlet transform. It is also planned to effectively combine face recognition technology with accurate human location technology to improve the effectiveness of post-disaster rescue.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TD76;TP391.41

【相似文献】