Target Detection Algorithm Based on Improved YOLOv11

To address challenges in balancing detection accuracy and speed, detecting small targets, and limited computational resources in complex environments, this paper proposes an improved algorithm named GCD-YOLO based on YOLOv11. The algorithm integrates depthwise separable convolution to reduce network size and computational complexity, incorporates a high-order attention mechanism (GSoP-Net) into the backbone network to enhance focus on target objects and contextual information, and employs stochastic value pooling to improve model robustness. The SIoU loss function is adopted to refine bounding box regression, while a deformable convolutionbased feature pyramid (DS-FPN) enables multi-scale feature fusion, leveraging information across dimensions to achieve precise detection of targets, particularly improving small-scale target detection. Experiments demonstrate that the improved algorithm achieves an mAP@ 0.5 of 96.9 %, a 6.8 percentage point increase over the baseline model, with inference speed maintained at 77 FPS (28 % faster than the baseline) and model parameters increasing by only <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4. 3 \%}$</tex>. Additionally, the algorithm reduces missed and false detections for small targets and is more deployable on resource-constrained devices.