Research on lightweight small target detection algorithms based on scale awareness

To address the challenge of detecting small targets in occluded, low-light, and other challenging environments, this paper improves and optimizes the YOLO11 model, proposing a Lightweight Multi-Scale YOLO (LMS-YOLO) model tailored for small target detection. First, a Multi-Scale Information Enhancement Block (MIEB) is introduced to strengthen feature extraction capabilities. Next, a Multi-Scale Multi-Branch Auxiliary Fusion FPN (MS-MBAFPN) spatial pyramid structure is designed, incorporating bidirectional fusion and cross-layer fusion techniques to improve the correlation between features across layers. Finally, the Focaler-MPDIOU is proposed as a bounding box loss function to address the imbalance of small target samples. The model is trained and validated using the VisDrone2019 dataset. Experimental results demonstrate that the proposed model achieves an average accuracy increase of 3.2% compared to the baseline, with a 2.1% improvement in the [email protected] value.