基于混合像元分解的庐山及周边林地遥感图像分

由于传感器分辨率的限制以及地物的复杂多样性，混合像元普遍存在于遥感影像中，在一定程度上影响到地物提取精度。以江西省庐山及周边地区2019年4月9日的ETM+影像为例，使用线性混合模型和非线性的BP神经网络方法对图像进行混合像元分解，利用2019年5月的SPOT数据及其与ETM+影像融合的分类结果对分解结果进行验证。结果表明，非线性的BP神经网络分解精度高于线性分解精度。对比使用BP神经网络分解图像提取林地面积的精度提高了1%~5%。

Remote Sensing Image Analysis of Forest Land in Lushan Mountain and Its Surrounding Area Based on Mixed Pixel Decomposition

Due to the limitation of sensor resolution and the complexity and diversity of ground objects, mixed pixels generally exist in remote sensing images, which affects the accuracy of ground object extraction to a certain extent. In this paper, the ETM+ image in Lushan, Jiangxi Province and its surrounding areas on April 9, 2019 was taken as an example. The mixed pixel decomposition of the image was carried out by using the linear mixing model and the nonlinear BP neural network method. The decomposition results were verified by using the SPOT data in May 2019 and the classification results of the fusion with ETM+ image. The results showed that the decomposition accuracy of nonlinear BP neural network was higher than that of linear neural network. Compared with BP neural network, the accuracy of forest land extraction was improved by 1%-5%.