采用特征融合的紫檀属内相似树种识别方法研究

紫檀属中的木材有很多属于名贵木材，不同树种之间十分相似。传统的木材识别方法多以木材解剖学为主，通过观察木材的切片结构特征对木材的树种进行判断，这类方法虽有较高的识别精度，但是其识别工艺较为复杂而且技术难度也相对较高。与木材解剖学相对应的是利用图像信息或光谱信息的木材树种识别方法，该类方法虽具有较为简单的识别工艺，但是在对同属相似木材树种进行识别时，往往不能够取得较好的识别效果。提出了一种基于木材切面光谱特征和纹理特征相融合的木材树种识别方法，该方法不仅识别工艺简单、自动化程度高，而且具有较高的识别精度。首先通过数码相机和光谱仪采集木材切面的图像信息和光谱信息，然后分别使用纹理特征提取方法和光谱特征提取方法提取两类特征的特征向量，接下来使用基于典型相关分析的特征级融合方法将这两个特征向量进行融合，最后使用支持向量机对融合后的特征向量进行分类识别。为了验证方法的有效性，以市场中常见的5种紫檀属树种的三个切面为研究对象，对这些木材树种进行了识别。实验结果显示，单独使用纹理特征的识别正确率最高为80.00%，单独使用光谱特征的识别正确率最高为94.40%，使用融合的特征最高的识别正确率可达99.20%。还将这5种木材树种与其他30种木材树种进行了混合，混合后的木材样本数量可达1750。实验进一步显示，该方法可以对包含紫檀属在内的35种木材的树种进行识别，其正确率可达98.29%。综上所述，木材的纹理特征和木材的光谱特征可以有效的相互补充，从而进一步提高识别正确率。最后还用所提出的方法与目前主流的方法进行了比较，结果发现所述的木材树种识别方法高于目前主流方法。

Similar Wood Species Classification Within Pterocarpus Genus Using Feature Fusion

There are much rare wood in the Pterocarpus genus. Rosewood is very similar to different tree species. Traditional wood identification methods are mainly based on wood anatomy, and the wood species are judged by observing the structural characteristics of wood slices. Although this method has a high identification accuracy, its identification process is relatively complex, and the technical difficulty is relatively high. Corresponding to wood anatomy is the identification method of wood tree species using image or spectral information. Although this kind of method has a relatively simple identification technology, it often fails to achieve a good identification effect in identifying similar wood species belonging to the same genus. This paper proposes a wood species identification method based on the fusion of spectral features and texture features of wood section. This method has a simple identification process, a high degree of automation, and a high identification accuracy. First collected by digital camera and a spectrometer wood, slice image information and spectral information, and then respectively using texture feature extraction method and spectrum feature extraction method to extract the characteristics of two kinds of the feature vector, then using the feature level fusion method based on canonical correlation analysis to these two characteristics vector fusion, finally using support vector machine (SVM) for the fusion of feature vector classification recognition. In order to verify the effectiveness of the method, three sections of 5 species of Rosewood species commonly found in the market were taken as research objects to identify these wood species. The experimental results show that the highest recognition accuracy is 80.00% when texture features are used alone, 94.40% when spectral features are used alone, and 99.20% when fused features are used. In this paper, these 5 wood species were mixed with 30 other wood species, and the number of mixed wood samples could reach 1 750. The experimental results show that the method can identify 35 wood species, including Rosewood, and the accuracy rate is 98.29%. To sum up, the texture features and spectral features of wood can effectively complement each other to further improve the recognition accuracy. At the end of this paper, the proposed method is compared with the current mainstream method, and the results show that the wood species identification method described in this paper is higher than the current mainstream method.