Abstract: A model for infrared–visible object detection that integrated global and local features was developed to improve the performance of multimodal object detection in complex scenes. The model was designed using a multisource heterogeneous network architecture. For visible–light images, group convolution based on feature importance was used to build a convolutional neural network, extracting the detailed texture features of objects from local spatial dimensions through a sliding window mechanism. For infrared images, a highly efficient multihead self-attention mechanism was employed to build a visual transformer network, capturing the salient features of objects through global context modeling. To facilitate the interaction of infrared and visible features, a dynamic weight-adaptive fusion strategy was designed to achieve cross-modal feature complementarity through key-feature enhancement and modal-weight allocation. Additionally, a spatial-channel multiscale collaborative optimization module was developed to alleviate information conflicts when fusing features of different scales. The most effective features were highlighted and noise information from the spatial and channel dimensions was suppressed through an attention mechanism, thereby reducing the mutual interference between features. Experiments on multiple standard datasets showed that the proposed method significantly improved multimodal feature extraction, cross-modal feature interaction, and multiscale feature fusion compared to the YOLOv10 baseline model. Moreover, compared to similar existing networks, this method demonstrated better detection performance, enabling accurate and efficient object detection in complex environments.