Improved YOLO11 with multi-scale feature enhancement for indoor small object detection

Abstract

Indoor small object detection is a core technology for smart home systems. The original YOLO11 model suffers from small-target feature loss, poor multi-scale feature fusion, and frequent missed detection in complex indoor scenes. To address these issues, this paper proposes an improved YOLO11 method for indoor small object detection. We replace partial C3k2 downsampling with a 5 × 5 large convolution kernel (stride=2) to expand the receptive field and retain small-target details. The SPPF module is substituted with AIFIv8 to enhance high-level semantic feature fusion. We further introduce the HFFE hierarchical feature fusion module to strengthen cross-scale feature interaction and upgrade the detection head to Detect_Efficient for more accurate prediction. Meanwhile, the Focaler-IoU-DIoU loss is adopted to optimize bounding box regression for small targets. Experiments on the VOC2007 dataset show that our method improves mAP@50%by 3.9% and small-target AP by 4.1% compared with the baseline YOLO11n, demonstrating better performance for indoor multi-scale small object detection.