Denoising Algorithm of LJ1-01 Nightlight Data with Hybrid Chi-square Distribution

摘要

To distinguish the effective light points caused by human activities from the noise points, and improve the extraction accuracy of feature targets, a denoising algorithm of LJ1-01 nightlight with hybrid Chi-square distribution was proposed. It was assumed that LJ1-01 data and noise subsets are superpositions of multiple Chi-square. and based on the principle of least squares, the indirect adjustment equation was constructed by combining effective statistical lighting value, and the proportion of each superposition state was solved. Next, a continuous probability density curve with sharp fitting accuracy was constructed based on the proportion. Finally, the abundance function was constructed. According to the offset of weight centers of different noise subsets in the original image, noise was removed. To verify the universality and robustness of the algorithm, four regions were selected as experimental objects in China according to the reflective characteristics of discrete objects. The feature similarity of the images after noise elimination is higher than 0.82 and the structure similarity is higher than 0.94. Meanwhile, three regions in Beibu Gulf were selected to extract the number of ships to verify the denoising effect. The results demonstrate that the accuracy of ship extraction using hybrid Chi-square denoising algorithm is above 90%.