IHS变换与低通滤波相结合的遥感图像增强模型

IHS transform and low pass filter based image enhancement for crop classification

利用遥感图像分类提取作物面积时,由于作物长势的差异,同种作物往往表现出不尽相同的图像光谱特征,在大片连续或近似连续图斑中出现光谱突变,类似于图像噪声。这种微小的差异使得图像分类时同种作物容易被误判为其他地物,增大了分类后处理的难度,降低了分类精度。作者研究的SMM模型首先对原始多光谱图像做IHS变换,将原始图像分离成I(亮度)、H(色度)和S(饱和度),然后对色度H和饱和度S进行卷积滤波运算,得到H′和S,′再将I、H′和S′做IHS逆变换,得到新图像。SMM模型旨在通过图像平滑解决目前图像非监督分类方法上存在的不足,但又在图像平滑的同时保留了原始图像的空间分辨率。通过分类试验验证,使用SMM模型进行图像增强,可以提高图像分类的精度。

One of the main task in agricultural monitoring using remote sensing is to estimate crop area through image classification. But sometimes the classification results are not satisfactory resulting from the spectral difference of the same variety of crops for their different growth conditions. In the image the spectral difference of the same crops is like noise and gets more difficult in post classification processing. An image smoothing model （SMM） is developed to improve the clasification. SMM first transforms the multispectral image into intension （I）, hue（H） and saturation（S）; then a convolution filter is used only on its hue and saturation to get new hue （H＇） and saturation （S＇） components. The original I, the new H＇and S1are together used to get a new RGB image by inverse IHS transformation. Therefore, SMM model preserves the spatial resolution when smoothing a multispectral image. The enhanced image using SMM model proves to be able to obtain higher accuracy of classification.