Low-light image enhancement improves visibility and detail in poorly lit conditions, crucial for applications in security, medical imaging, and autonomous systems, but existing methods face many challenges, including noise and color distortion, complex multi-stage training process, and insufficient modeling of long-range dependencies. The noise and color distortion problems arise because existing methods fail to effectively suppress noise and maintain color consistency during the brightening process. The complex multi-stage training process not only increases the training time and computational cost, but also leads to error accumulation between different stages. Moreover, current techniques often struggle to effectively capture long-range interactions and the inherent self-similarity found in different regions of images, which affects the processing effect of complex low-light images. We propose an innovative method to address these problems through a decomposition and reconstruction framework of illumination and reflectance components, supplemented by Fourier-guided component prediction (FGIDRCP). This method simplifies the training process and improves efficiency, and uses a signal-to-noise ratio (SNR) dynamic weight adjustment method to adaptively handle noise and details in different regions. First, our model predicts the illumination difference and reflectance components to generate a preliminary normal-light image, which is then refined and denoised to finally generate the enhanced image. Through Fourier-guided component prediction, we are able to better extract and utilize the structure and texture information of the image, helping the network to decompose the reflectance component and thus reduce image distortion. The SNR-guided dynamic weight adjustment module optimizes denoising performance by adapting to varying noise levels and improving image clarity. Experimental results show that FGIDRCP performs well in low-light image enhancement tasks.