基于多尺度LBP特征融合的遥感图像分类

Classification of remote sensing images based on multi-scale feature fusion using local binary patterns

针对高光谱遥感图像分类问题,传统的特征提取方法常忽略其本征属性信息和图像的多尺度局部结构特性而使其获取的图像信息量较少,为改进这一缺陷,提出了一种多尺度灰度和纹理结构特征融合的方法模型(multi-scale gray and texture structure feature fusion,Ms_GTSFF)进行遥感图像特征提取。首先用多尺度方法提取图像不同尺度下的灰度属性特征,然后利用局部二进制模式的思想获得图像的局部纹理特征信息,同时利用多尺度还能够获取图像更大感受野的特征,接着利用得到的多尺度LBP直方图获取每种编码所对应的灰度属性信息,最后将上述得到的多尺度特征信息进行编码融合,构成了Ms_GTSFF特征提取模型,再连接多种机器学习分类器进行分类识别。以雄安新区(马蹄湾村)航空高光谱遥感影像作为测试数据集,对数据分块预处理后再进行特征提取与分类测试,最高获得了99.44%的分类准确率,在遥感图像分类上与传统方法的识别能力相比有很大的提升,验证了提出模型对于增强遥感图像的特征提取能力以及提高分类识别性能的有效性。

For the classification of remote sensing images,traditional feature extraction methods frequently ignore their intrinsic properties and the multi-scale local characteristics of the images.As a result,only a small amount of image information can be acquired.Given this,this study proposed a model of multi-scale gray level and texture feature fusion(Ms_GTSFF)for the feature extraction of remote sensing images,and the extraction steps are as follows.Firstly,extract the gray-level features of the images at different scales.Then obtain the local texture features of the images using the local binary pattern(LBP)algorithm and meanwhile,obtain the image features of a larger receptive field using a multi-scale method.Afterward,obtain the gray-level attributes corresponding to various codes using the obtained multi-scale LBP histograms.Finally,code and fuse multi-scale feature information obtained from the above steps to constitute the Ms_GTSFF feature extraction model,to which multiple machine learning classifiers are connected for classification and recognition.Taking the aerial hyperspectral remote sensing images of Xiongan New Area(Matiwan Village)as the test dataset,the feature extraction and classification tests were performed following the data preprocessing by blocks.The classification accuracy was up to 99.44%,indicating a great improvement in the recognition capability compared with traditional methods.This verified the effectiveness of the proposed model in enhancing the feature extraction capability and improving the classification and reorganization performance of remote sensing images.