Underwater images suffer from uneven illumination, severe color cast, and the scarcity of high-quality labeled data. To address these issues, a semi-supervised underwater image enhancement method based on adaptive illumination correction is proposed. The method adopts a teacher-student learning framework, in which a small amount of labeled data and a large amount of unlabeled data are jointly utilized for training. An adaptive illumination correction module is introduced to achieve differentiated compensation for regions with varying brightness through illumination component separation and adaptive adjustment, thus alleviating color distortion and uneven local exposure. In addition, a pseudo-label quality assessment mechanism is introduced to dynamically select reliable unlabeled samples, optimizing the training process. Experimental results show that, on public datasets such as UIEB, the proposed method outperforms existing mainstream methods on multiple no-reference evaluation metrics, including UIQM, UCIQE, and URanker, and exhibits excellent performance in color restoration and detail preservation. When deployed in real-world scenarios, the model still effectively enhances both subjective and objective image quality, demonstrating good practicality and generalization ability.