2. 95806 部队，北京100071;

3. 93116 部队，沈阳110141;

4. 31693 部队，哈尔滨150036

法由空间特征提取、多粒度扫描和深度森林分类三部分组成。首先提取高光谱影像的形态学属性剖面

特征，将所提取的特征与原始光谱特征进行拼接获得融合后的空--谱特征; 然后通过多个尺度的滑动

窗口对空--谱特征进行多粒度扫描，以实现特征重用; 最后将多粒度扫描后的特征输入到深度森林分

类器中进行分类。采用Pavia 大学高光谱数据集和Salinas 高光谱数据集进行分类试验，所设计的分类方法在两组高光谱数据上分别取得了98. 44% 和98. 47% 的总体分类精度。试验结果表明所设计的分类方法能够有效地提高高光谱影像的分类精度。

2. Troops 95806，Beijing 100071，China;

3. Troops 93116，Shenyang 110141，China;

4. Troops 31693，Harbin 150036，China

spatial feature extraction，multi-grained scanning and deep forest classification． Firstly，the morphological attribute profile features of hyperspectral images are extracted，and the extracted features are concentrated with the

original spectral features to generate the spatial-spectral features; secondly，the obtained spatial-spectral features are scanned with multiple scale sliding windows to achieve feature reuse; finally，the features after multi-grained scanning are input into the deep forest classifier． The classification experiments are carried out with the Pavia University and Salinas hyperspectral data set． The overall classification accuracy of the designed classification method  is 98. 44% and 98. 47%，respectively on the two hyperspectral data sets． The experimental results show that the

designed classification method could effectively improve the classification accuracy of hyperspectral images．