基于分块压缩感知的大数据量遥感图像薄云去除方法研究

摘要:

大数据量遥感图像薄云的存在影响图像的清晰度,受到薄云分布不均匀性和随机性影响,其采样存在信息不完整和噪声干扰问题,使得特征不完整,影响图像透射率,导致图像薄云特征分析不准确,薄云去除效果差。为此,提出基于分块压缩感知的大数据量遥感图像薄云去除方法。先定义像素感知对象的分块矩阵,基于分块压缩感知算法计算采样峰值的信噪比参量,实现大数据量遥感图像采样,以解决采样信息不完整,存在噪声干扰的问题。然后利用所得采样结果,求解图像的空间特征、灰度剖面图特征与频率特征,完成大数据量遥感图像薄云特征分析,提升特征分析效果。最后参考优化去除因子与导向滤波优化透射率,改进大气光值,实现对薄云去除参量的分波段迭代,完成遥感图像薄云去除设计。实验结果表明,应用所提方法,可缩小薄云覆盖区域与其边缘区域的像素差值,使薄云区域的色温水平更接近于整幅图像的色温均值,提高遥感图像清晰度,应用效果较好。

Abstract:

The presence of thin clouds in large-scale remote sensing images affects the clarity of the images, and is affected by the uneven and random distribution of thin clouds. The sampling process suffers from incomplete information and noise interference, resulting in incomplete features and affecting image transmittance, leading to inaccurate analysis of thin cloud features and poor removal of thin clouds. Therefore, a method for removing thin clouds from large-scale remote sensing images based on block compression perception is proposed. Firstly, define the block matrix of the pixel perception object, calculate the signal-to-noise ratio parameter of the sampling peak based on the block compression perception algorithm, and achieve large-scale remote sensing image sampling to solve the problem of incomplete sampling information and noise interference. Then, using the obtained sampling results, the spatial features, grayscale profile features, and frequency features of the image are solved to complete the analysis of thin cloud features in large-scale remote sensing images and improve the effectiveness of feature analysis. Finally, referring to the optimization of removal factors and guided filtering optimization of transmittance, the atmospheric light value is improved, and the sub-band iteration of thin cloud removal parameters is achieved to complete the design of remote sensing image thin cloud removal. The experimental results show that the proposed method can reduce the pixel difference between the thin cloud coverage area and its edge area, making the color temperature level of the thin cloud area closer to the average color temperature of the entire image, improving the clarity of remote sensing images, and achieving good application effects.