选煤厂工况环境下人脸识别方法研究

Research on face recognition method in working environment of coal preparation plant

针对选煤厂人脸图像信息易受复杂环境因素影响导致识别难度较大的问题,研究了一种选煤厂工况环境下的人脸识别方法。对归一化的选煤厂原始人脸图像进行Gabor小波变换,得到8个方向、5个尺度下的特征图谱;用改进AR-LGC编码算法进行编码,并对编码后同一尺度下不同方向的图谱进行特征融合,得到图像的融合特征图;将融合特征图划分为多个子块,统计分块直方图并加权级联得到直方图特征向量,将特征向量送入残差神经网络中训练,实现对选煤厂人员的人脸识别。改进AR-LGC编码算法增强了选煤厂人脸图像纹理相关度,解决了图像纹理相关度不足的问题,在弱化干扰特征的同时,保留了人脸图像中更多重要特征,缓解了人员面部受煤灰污染的问题。实验结果表明:当选煤厂人脸受到煤灰污染时,采用改进AR-LGC编码算法提取的特征保留了局部特征粗粒度,具有较好的抗噪性;本文方法的识别率为94.5%,平均耗时为0.9330 s,与同类算法相比,在牺牲部分时间性能的条件下提升了识别率,牺牲的时间性能在可接受范围内。

The face image information of coal preparation plant is easily affected by complex environmental factors,which makes the recognition difficult.In order to solve this problem,a face recognition method in working environment of coal preparation plant is proposed.The Gabor wavelet transform is applied to the normalized original face image of coal preparation plant to obtain characteristic maps of 8 directions and 5 scales.The method encodes with the improved AR-LGC coding algorithm,and characteristic fusion is performed on the encoded maps of different directions at the same scale to obtain the fused characteristic map of the image.The fused characteristic map is divided into multiple sub-blocks,and the histogram characteristic vector H is obtained by counting the block histogram and weighting cascade.Then H is trained in residual neural network to realize the face recognition of the personnel in the coal preparation plant.The improved AR-LGC coding algorithm enhances the texture correlation of face images in coal preparation plant,solves the problem of insufficient image texture correlation,retains more important characteristics in face images while weakening the interference characteristic,and alleviates the problem of personnel faces being polluted by coal ash.The experimental results show that when the faces of coal preparation plant are polluted by coal ash,the characteristics extracted by the improved AR-LGC coding algorithm retain the local characteristics coarse granularity and have better noise immunity.The recognition rate of the improved AR-LGC coding algorithm is 94.5%and the average time consumed is 0.9330 s.Compared with similar algorithms,the recognition rate of this algorithm has improved under the condition of sacrificing part of the time performance,and the sacrificed time performance is acceptable.