<p>Image dehazing technology aims to restore clear and realistic images from those degraded by haze, and is widely applied in fields such as remote sensing image processing, security surveillance, and aerospace. However, existing dehazing methods have not fully utilized the diverse inputs provided by image enhancement techniques, which limits the optimization potential of the models. Additionally, most contrastive learning-based dehazing methods rely on a single type of degraded sample, which hinders effective learning of haze characteristics across different scenarios. Furthermore, traditional contrastive loss functions typically only compare hazy and clear images directly, with a relatively simplistic comparison standard that restricts model performance. To address these issues, we propose a semi-supervised dehazing method based on image enhancement and multi-negative contrastive auxiliary learning (IE-MNCAL), built on the mean teacher model. This method preprocesses the input hazy images using image enhancement techniques such as histogram equalization, white balance, and gamma correction. The enhanced images are then concatenated with the original hazy images along the channel dimension, providing more diverse input features for the dehazing model. To further improve model performance, we construct an auxiliary negative sample set that stores both the preprocessed and original hazy images, thereby increasing the variety of degraded samples. Based on this, we introduce a multi-negative contrastive auxiliary loss function, which uses multiple types of degraded images from the auxiliary negative sample set to reduce the distance between the anchor point and positive samples from various directions. This effectively enhances the generalization ability of the dehazing model. Experimental results show that our proposed dehazing method, IE-MNCAL, effectively removes haze in various scenarios across multiple datasets and outperforms state-of-the-art dehazing methods in terms of dehazing quality.</p>