Existing dehazing methods based on classical theory often focus on two problems: the estimation of the transmission and ambient light, which is prone to suffering from blurry image along haze boundaries. To address the issue, we propose a novel network based on correcting kernel-based for single-image haze removal(CKNet), which gradually improve the performance of removing haze by a kernel correct (KC)block designed. We first construct a U-like framework for fog removal network, then combine it with a kernel correction network constructed using small convolutional neural networks to predict the kernel parameters more suitable for the fog scenario, and finally form the optimal network parameters after continuous iterations. Extensive experimental results show the effectiveness of our method compared with other state-of-the-art methods based on deep leaning and monochromatic atmospheric scattering model.