The work aims to propose the Retinex low light image enhancement algorithm based on multi-scale adaptive NSST, regarding the problem that the traditional frequency domain method cannot properly keep the high-frequency details of the image due to the insufficient scale in low light image enhancement. Firstly, after the low light image was transformed into HSI color space, I channel was processed separately to achieve the fidelity effect of image color information. Then, the reflection component of I channel was obtained by Retinex algorithm, so as to remove the influence of illuminance component on image brightness. After gamma adjustment of the reflection component, based on La (average brightness), Pa (average contrast) and Ia (information entropy), the best high frequency component was obtained by the adaptive NSST decomposition. In subjective observation and objective non reference image quality evaluation, the enhancement effect and evaluation score of the proposed algorithm were better than other algorithms. After the optimization of adaptive parameters, the contrast of low light image is improved, and the visibility and image quality are improved.