Images taken in low-light conditions have a narrow dynamic range and darker tones, making them difficult to see with the human eyes. The Discrete Wavelet Transform (DWT) is reversible, thus enabling the image to be broken down into sub-bands without loss of information, minimizing redundancy. In this paper, we propose sub-band enhancement of low-light images based on a residual diffusion model with discrete wavelet transform. We use discrete wavelet transforms to achieve contrast enhancement. We combine the residual diffusion model to realize the conversion of low-light images to normal lighting. First, we decompose the input image into LL, LH, HL, and HH sub-bands to obtain low- and high-frequency components. Secondly, we use residual diffusion model to generate LL sub-band corresponding to normal light images from the LL sub-bands of low-light images, and use a CNN transfer network with attention mechanism to convert LH, HL and HH sub-bands. Experimental results show that our method enhances low-light images, as well as outperforms state-of-the-art (SOTA) methods in visual quality and quantitative measurements.