Traditional Chinese Medicine (TCM) has gained prominence in clinical practice, with tongue diagnosis, a key technique, now being integrated with Artificial Intelligence (AI) to achieve more objective and quantifiable results, thereby mitigating reliance on subjective judgment. However, challenges such as poor lighting conditions and limited imaging equipment often compromise image clarity, complicating tongue detection and identification. To address these issues, we propose a Dual-Task Feedback Learning (DTFL) framework, designed to enhance tongue detection in patient images by improving image quality. In our approach, Super-Resolution (SR) serves as a preliminary task preceding Tongue Detection (TD), enabling the TD network to process high-quality images for more accurate results. To further improve the interaction between SR and TD tasks, we incorporate Feature Alignment (FA) loss, which establishes a feedback connection that allows the SR network to acquire task-specific knowledge from the TD network. Additionally, we introduce a quality fusion augmentation and alternate training strategy to address potential challenges associated with FA loss during training. To the best of our knowledge, we are the first to integrate SR into TD. Experiments demonstrate that DTFL significantly improves performance by generating SR images that are optimally suited for TD.