<p>To address drastic scale variations, dense distribution of small targets and complex backgrounds in remote sensing images, the trade-off between network lightweight design and detection accuracy has become a key bottleneck in remote sensing object detection. These issues greatly hinder intelligent interpretation and automatic analysis of massive remote sensing data, so it is urgent to develop high-performance lightweight detection methods suitable for complex remote sensing scenarios. Accordingly, this paper proposes an improved YOLOv8 lightweight detail dynamic fusion algorithm (LDDF-YOLO) for remote sensing object detection, to remedy the insufficient adaptation capability of vanilla lightweight YOLOv8 in remote sensing object detection tasks. First, the Spatial Pyramid Pooling-Fast (SPPF) module at the backbone is replaced with a combined structure consisting of a convolution layer and the Attention-based Intra-scale Feature Interaction (AIFI) module. It enhances cross-scale global feature fusion via the self-attention mechanism, effectively suppresses background interference, highlights spatial boundary features between targets and backgrounds, and improves the recognition capability of small objects under complex backgrounds. Second, a Dynamic Adaptive Focus Pyramid Network (DAFPN) is constructed to restructure the neck structure. It adopts a dynamic weighting strategy to efficiently fuse multi-scale features from deep, middle and shallow layers, which effectively alleviates detection difficulties caused by scale variation, and raises feature utilization while retaining detailed feature information. Finally, a Lightweight Shared Detection Head (LSDH) is designed. Combined with multi-directional detailed texture feature extraction and shared convolution mechanism, it reduces the parameter volume by 65.56% compared with the original detection head, realizing a lightweight design without sacrificing detection accuracy. Experimental results on the NWPU VHR-10 dataset show that compared with YOLOv8n, LDDF-YOLO increases recall and mAP50 by 7.9% and 5.2%, and cuts down model parameters and computational complexity by 40.2% and 27.2%, which verifies its excellent balance between lightweight design and detection accuracy. On the RSOD dataset, its mAP50 rises by 3.6%. On the VisDrone2019 dataset with dense small objects and complicated backgrounds, its mAP50 is improved by 1.2%, and it also achieves excellent inference speed on the VisDrone2019 test set. These results confirm that LDDF-YOLO achieves advanced performance in lightweight design, detection accuracy and inference speed, and possesses robust detection performance in challenging remote sensing scenarios with dense small targets and complex backgrounds.</p>