Objective Remote sensing image target detection plays a crucial role in resource exploration, urban planning and management, and disaster emergency response. However, targets in remote sensing images are typically characterized by small size, large scale variations, complex backgrounds, blurred edges, and insufficient texture information. These challenges are further exacerbated in low-resolution images, where the distinction between targets and the background becomes less pronounced. Although significant progress has been achieved in remote sensing target detection, small target detection still suffers from missed detections and limited feature representation. Therefore, this study aims to improve the accuracy and robustness of small target detection in remote sensing images while enhancing training efficiency and localization precision.

Methods To address the issues of unclear features, complex backgrounds, and densely distributed small targets, a remote sensing small target detection algorithm integrating attention mechanisms and feature enhancement is proposed. First, a multi-level feature collaboration strategy is constructed by combining channel attention and spatial attention mechanisms with a gating mechanism to dynamically regulate information flow, thereby strengthening cross-level feature interaction and suppressing background noise. Second, a multi-scale feature enhancement module is designed by integrating three-dimensional convolution and multi-scale feature encoding to form an enhanced neck network, which improves the feature extraction and representation capabilities for small targets. Finally, SIoU and NWD are jointly employed as regression loss functions to optimize bounding box matching, accelerate convergence, and enhance localization accuracy.

Results and Discussions Experiments conducted on the NWPU VHR-10, DIOR, and RSOD datasets demonstrate the effectiveness of the proposed method. Compared with the original baseline model, the improved algorithm achieves improvements in mAP@0.5 of 9.9%, 3.1%, and 3.2%, respectively. Visualization results further indicate that the proposed method can successfully detect small targets that were previously missed, particularly in complex scenes with densely distributed targets. These results verify that the integration of multi-level attention collaboration and multi-scale feature enhancement effectively improves small target representation and reduces missed detections.

Conclusions The proposed attention- and feature-enhancement-based small target detection algorithm significantly improves detection accuracy and localization performance in remote sensing images. By enhancing cross-level feature interaction and strengthening multi-scale representation, the method effectively addresses the challenges of small target detection in complex scenarios. The experimental results confirm its robustness and superiority over the baseline model, demonstrating its potential for practical remote sensing applications.