The hyperspectral remote sensing data contains rich information of reflective spectrum of surface feature,however,the original reflection data includes the massive noises,which affects the absorption feature of reflective spectrum and degrades the data analysis precision greatly. The study of hyperspectral remote sensing data noise filtering algorithm is the key to improve data analysis. The hyperspectral image noise filtering technology is studied. Deep research in high-frequency characteristics of pushbroom hyperspectral imager (PHI) stripe noise,in view of the disadvantages of currently common and several improved moment matching methods,an improved smooth filtering algorithm by row is proposed. The line-average curve of the bands contained strip noise is smoothed and gray value of each pixel in the image is also adjusted to reduce the gray differences between the lines. The peak signal-to-noise-ratio of the gained image has been improved and better effect is got comparing with moment matching method by bands. While strip noise is weakened well,the radiative feature of original image is retained.