基于I-BGLAM纹理和光谱融合的高光谱显微成像木材树种分类

为了提高木材树种分类的正确率，提出了一种基于I-BGLAM纹理特征和光谱特征融合的高光谱图像的木材树种分类方法。实验数据是利用SOC710VP高光谱成像仪获取的可见光/近红外（372.53~1 038.57 nm）范围内的高光谱图像。首先，利用基于OIF的特征波段选择方法降低高光谱图像的维数，选择出含有信息量大的波段。其次，对选择出的波段图像使用NSCT及NSCT逆变换得到融合图像，对得到的融合图像使用I-BGLAM提取其纹理特征。与此同时，对高光谱图像的全波段求取平均光谱并进行S-G（Savitzky-Golay）平滑得到光谱特征。最后，将得到的纹理特征和光谱特征融合后送进极限学习机（ELM）中进行分类。此外，还和基于灰度共生矩阵（GLCM）的木材识别的传统方法以及近几年木材树种识别领域内被提出的主流方法进行了比较。该研究主要创新点有两个：一是将强纹理提取器I-BGLAM用于高光谱图像中提取其纹理特征；二是提出一种新的特征融合的模型用于高光谱图像的分类。针对8个树种的实验结果表明，单独使用I-BGLAM提取的纹理特征来进行分类的正确率最高可到达88.54%，而使用GLCM提取纹理特征的传统方法正确率最高只有76.04%，该结果可以得出本文使用I-BGLAM在纹理特征提取方面要优于GLCM，这为后面建立的融合模型打下很好的基础，单独使用平均光谱特征来分类的正确率最高可以达到92.71%，使用所提出的特征融合方法所得到的分类正确率最高可达到100%，这说明使用所提出的融合模型来分类要比以前单独使用某一种特征的分类模型要好。此外，使用所提出的方法得到的分类正确率要高于本领域内其他两种主流的识别方法。因此，所提出的基于I-BGLAM纹理特征和光谱特征融合的方法能够提高木材树种分类的正确率，该方法在木材树种分类方面有着一定的利用价值。

Wood Species Classification With Microscopic Hyper-Spectral Imaging Based on I-BGLAM Texture and Spectral Fusion

To improve the accuracy of wood species classification,a method is proposed based on I-BGLAM(Improved-Basic Gray Level Aura Matrix)texture features and spectral features fusion in this paper.Experimental data are hyper-spectral images in the visible and near-infrared spectral band(i.e.,372.53~1038.57 nm)obtained by SOC710VP hyper-spectral imaging system.Firstly,the feature band selection method based on OIF(Optimum Index Factor)was used to reduce the dimension of hyper-spectral images and select the band containing a large amount of information.Secondly,NSCT(Nonsubsampled Contourlet Transform)and inverse transformation of NSCT were used to obtain the fusion image for the selected band images,and I-BGLAM was used to extract its texture features for the obtained fusion image.At the same time,the average spectrum of the whole band of hyper-spectral image was obtained,and the spectral characteristics were obtained by S-G(Savitzky-Golay)smoothing.Finally,the obtained texture features and spectral features were fused and sent to ELM(Extreme Learning Machine)for classification.In addition,the method proposed in this paper is compared with the traditional method of wood identification based on GLCM(Gray Level Co-occurrence Matrix)and the mainstreams method proposed in the field of wood species identification in recent years.There are two main innovations in this paper.One is to use the strong texture extractor I-BGLAM to extract its texture features from hyper-spectral images;the other is to propose a new feature fusion model for the classification of hyper-spectral images.The experimental results of 8 tree species show that the accuracy of using I-BGLAM to extract texture features for classification was up to 88.54%,while the accuracy of using GLCM to extract texture features was up to 76.04%.The results show that the use of I-BGLAM in this paper is better than that of GLCM in texture feature extraction,which lays a good foundation for the fusion model established later.The accuracy of classification by using the average spectral features alone can reach 92.71%.The classification accuracy of the proposed feature fusion method can reach up to 100%.This shows that it is better to use the fusion model proposed in this paper for classification than to use the classification model of a single feature.In addition,the classification accuracy obtained by using the method proposed in this paper is higher than the other two mainstream recognition methods in this field.Therefore,the method proposed in this paper based on I-BGLAM texture feature and spectral feature fusion can improve the accuracy of wood species classification,which has certain utilization value in the classification of wood species.