MSF-DDN: A Multi-Modal Dynamic Fusion Framework for Robust Target Detection in Smoke-Degraded Environments

With the rapid development of computer vision and artificial intelligence technology, target detection technology shows great application potential in the fields of security monitoring and automatic driving. However, in complex environments, such as in the presence of smoke, uneven illumination, and other harsh conditions, the performance of traditional target detection algorithms is seriously affected, and the single-modal data can hardly reflect the target features comprehensively, resulting in limited accuracy and generalization ability of target detection. To address the above challenges, this paper proposes a method based on Multi-modal Synergistic Fusion with Dynamic Detection Network (MSF-DDN). The method firstly matches and aligns the features of visible data and infrared spectral data by a ternary weighting algorithm and then utilizes the Laplace pyramid method to perform a weighted fusion of multi-modal data and realizes the smoke-free image restoration by a de-smoke algorithm. In the target detection module, dynamic convolution, a deformable component model, and a two-layer routing attention module are introduced to improve the YOLO11 model, which effectively improves the accuracy and generalization ability of target detection. The experimental results show that compared with the traditional methods, MSF-DDN exhibits higher target detection accuracy and robustness in complex environments. Especially in the scenario with smoke, MSF-DDN's Precision, Recall and F1-Score are significantly better than traditional models such as YOLO11 and YOLO8. This study provides new ideas and methods for target detection in complex environments and provides solid technical support for the development of intelligent security and automatic driving.