Research on Multi-modal Remote Sensing Image Object Detection Based on YOLOv11-AM

For the purpose of securing strategic advantages in fields such as battlefield reconnaissance, strike situation assessment and border monitoring, object detection in remote sensing images proves to be irreplaceably significant. However, traditional recognition methods have non-negligible limitations for both visible light-based and infrared-based remote sensing images. Environmental lighting conditions tend to exert an impact on visible light images, while infrared images are frequently characterized by the absence of rich texture information. To address these challenges, we have improved a multi-modal object detection model YOLOv11-AM that integrates convolution and attention mechanisms. Based on the latest YOLOv11 structure, collaborative data within visible light and infrared images is aimed to be fused, so as to boost the efficiency of target recognition in remote sensing images. To address the need for comprehensive feature extraction of various modalities, we incorporate a convolution-attention fusion module as an auxiliary tool Secondly, a differential fusion module is employed to fuse the acquired features, so as to achieve deeper multi-modal information fusion. We conducted experimental verification using the public remote sensing dataset VEDAI, which proves that the model can deeply fuse information from different modalities, thus effectively improving the accuracy of object detection.