基于多融合多尺度特征的高光谱图像分类研究

针对高光谱图像(HSI)波段之间的冗余性给高光谱图像分类结果产生的不利影响,研究基于多融合多尺度特征的高光谱图像分类方法。将采用于主成分分析降维处理的HSI数据作为多尺度特征多融合残差网络输入,利用多尺度特征多融合残差块提取HSI中的光谱特征和空间特征,并组成若干组光谱-空间特征;采用支持向量机展开分类处理,获取各光谱-空间特征的概率输出结果和权重,建立多特征加权概率融合模型,利用最大后验概率获取高光谱图像分类结果。实验结果表明:光谱-空间多尺度特征融合残差块数量为2+2模式、空间输入尺寸大小为9×9,可获取最佳多尺度特征融合残差网络;所提方法抗噪能力较好,可较好体现地物细节信息;且具备较高的高光谱图像分类精度。

Hyperspectral image classification based on multi-fusion and multi-scale features

Aiming at the redundancy’s adverse effect between the bands of hyperspectral images(HSI)on the classification results of hyperspectral images,a hyperspectral image classification method based on multi-fusion multiscale features was studied.The HSI data used in principal component analysis dimensionality reduction processing were used as the input of the multi-scale feature multi-fusion residual network,and the multi-scale feature multi-fusion residual block was used to extract the spectral features and spatial features in the HSI.Several groups of spectral-spatial features were formed.Support vector machine(SVM)was used to expand the classification process to obtain the probability output results and weights of each spectrum-space feature.A multi-feature weighted probability fusion model was established to obtain hyperspectral images’classification results with the maximum posterior probability.The experimental results show that:the number of spectral-spatial multi-scale feature fusion residuals is 2+2 mode,the spatial input size is 9×9,the optimal multi-scale feature fusion residuals network can be obtained.The proposed method has better anti-noise ability and can better reflect the detail information of the surface features.Moreover,it has a high classification accuracy of hyperspectral images.