基于改进HOG特征与LPP算法的木材缺陷识别  

Wood defect recognition based on improved HOG feature and LPP algorithm

作　　者：彭骞 张华[1,2] 任万春 刘城 PENG Qian;ZHANG Hua;REN Wanchun;LIU Cheng(School of Information Engineering,Southwest University of Science and Technology,Mianyang 621010,China;Key Laboratory of Sichuan Province for Robot Technology Used for Special Environment,Mianyang 621010,China)

机构地区：[1]西南科技大学信息工程学院,四川绵阳621010 [2]特殊环境机器人技术四川重点实验室,四川绵阳621010

出　　处：《传感器与微系统》2022年第5期123-126,共4页Transducer and Microsystem Technologies

摘　　要：在木材缺陷的识别中,由于噪声干扰导致识别率降低,因此提出一种新的木材缺陷识别算法。首先通过计算不同尺度下各个方向的梯度,利用正交分解将各个方向的梯度进行结合,得到改进的梯度方向直方图(HOG)特征,提高了HOG特征的鲁棒性;其次将改进HOG特征与局部二值模式(LBP)特征线性加权得到融合特征,弥补了HOG特征没有表征缺陷纹理变化的缺点;最后通过增加全局信息与监督信息改进局部保持投影(LPP)算法并对其降维,再引入支持向量机(SVM)对上述特征进行分类。实验结果表明:在高斯噪声的信噪比为60 dB的环境下该算法的识别率达到97.13%。In identification of wood defects,the recognition rate is reduced due to noise interference,so a new wood defect recognition algorithm is proposed.Firstly,by calculating the gradients of different directions under different scales,the improved histogram of gradient direction(HOG)features are obtained by combining the gradients of each direction by orthogonal decomposition,which improves the robustness of HOG features.Secondly,the improved HOG features and local binary pattern(LBP)features are linearly weighted to obtain fusion features,which makes up for the defect that HOG features do not represent the texture changes of defects.Then,the local preserving projection(LPP)algorithm is improved by adding global information and supervision information,and the dimension is reduced.Finally,support vector machine(SVM)is introduced to classify the above features.The experimental results show that the recognition rate of the algorithm reaches 97.13 % when the SNR of Gaussian noise is 60 dB.

关 键 词：机器视觉 木材缺陷识别 改进梯度方向直方图 局部二值模式 局部保持投影 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]