Abstract: Images captured under low-light conditions commonly suffer from color distortion, loss of detail, and noise interference, which severely degrade performance in downstream vision tasks such as autonomous driving, object detection, and face recognition. To address these challenges, we propose a low-light image enhancement method that integrates the visual heat conduction mechanism with frequency-domain modeling. This method leverages the visual heat conduction mechanism to achieve smoothing enhancement from local to global scales. Concurrently, it integrates a frequency-domain approach to selectively reinforce detail information. Ultimately, this strategy improves color authenticity while effectively recovering texture features degraded by noise. Specifically, we employ wavelet transform to perform information-preserving downsampling and decompose features into low- and high-frequency components, enabling targeted design for each subband.To effectively model both global and local information in low-light images, we propose a low-frequency enhancement block that employs parallel visual heat conduction and state space modules to specifically recover the low-frequency subband. Furthermore, to effectively recover edge and texture details embedded in the high-frequency subband while suppressing noise interference, we introduce a high-frequency enhancement block that leverages the enhanced low-frequency information to refine the high-frequency components, thereby accurately restoring genuine high-frequency details. Experimental results show that, compared with state-of-the-art method, the low-light images enhanced by our approach achieve better performance both in terms of visual quality and objective metrics—specifically, (on the LOL-v1 dataset, our method improves PSNR by 0.34 dB and SSIM by 0.10).