山区遥感图像的特征选择与检索研究

Research on feature selection and retrieval of mountain area remote sensing images

针对遥感图像特征存在冗余及图像检索效率低的问题,提出一种基于贡献值的有效特征选择及其分类检索方法。对高分一号(GF-1)河北省阳原地区的村落、山脊、水体遥感图像,对比Hu矩、小波分解和灰度共生矩阵特征,发现使用Hu矩较适用于识别村落,而对山脊和水体的识别率低;引入区分山脊和水体的颜色特征,并与Hu矩构成特征空间,实现山区遥感图像分类和检索。实验结果表明,该方法能够选取出图像特征中的主要分量,具有高效、实用、方便的特点,检索查准率达到92%,查全率达到88%。

According to the problem that remote sensing image features redundancy and low efficiency of image retrieval,a new method of image classification retrieval feature selection based on feature component contribution value is given in the paper.Features on the remote sensing images of village in Yangyuan of Hebei,mountain and water are studied from images of GF-1 by the Hu moment,wavelet decomposition and gray level co-occurrence matrix.It is found that the Hu moment is more suitable for identification of village than mountain and water compared with the other characteristics.Further,color feature is introduced to distinguish the mountain from the water.The remote sensing images classification and retrieval are realized in the feature space with Hu moment.The experimental results show that the most significant components of the image features can be extracted.It has the characteristics of high efficiency,practicality and convenience,in which the retrieval precision of 92% and recall of 88% are obtained.