<p>With the development of deep learning and the advancement of unmanned aerial vehicle (UAV) research technology, the application of combining deep learning with UAVs is becoming increasingly common. In the field of target detection, UAVs, with their characteristics of high speed, flexibility, and extensive range of activities, have become an ideal platform for target detection algorithms. This paper proposes a small target detection method based on visible light and infrared multi-modal fusion from the perspective of UAVs. Leveraging the insensitivity of infrared cameras to lighting conditions, advantages such as obtaining clear images under poor lighting or adverse weather conditions are utilized. By effectively integrating this advantage with information with visible light, more accurate target detection is achieved, overcoming the adverse effects of lighting weather conditions. The proposed method designs a feature grouping perception module to effectively address the challenge of insufficient feature extraction for small targets in the infrared modality. Spatial feature extraction modules are designed in the feature extraction process of each modality to effectively reduce background interference, enabling the model to focus on useful information. A modality fusion feature extraction module is designed in the modality fusion stage, enabling the model to dynamically learn the contribution of each modality to different tasks in the channel dimension. Through dynamic adjustment of fusion weights, the model focuses on more critical information. Comparative experiments with the VEDAI public dataset and five different models show that the proposed detection model achieves the best performance, with mAP@50 reaching 78.6% and mAP@50:90 reaching 48.2%.</p>