Multilevel receptive field expansion network for small object detection

Abstract

Small object detection remains a bottleneck because there is little visual information about them, especially in the deep layers. To improve the detection performance of small objects, here, Swin Transformer is introduced as the model backbone network to extract rich features of small objects. Then, a multilevel receptive field expansion network (MRFENet) is proposed based on the characteristics of different stages in the Swin Transformer. Specifically, a receptive field expansion block (RFEB) is designed to acquire contextual cues and extract detailed information. The RFEB is carefully designed to target the required receptive fields of different layers and further refine the features. MRFENet combined with RFEBs implements the retention of small object context cues and the acquisition of receptive fields for the adaptive detection tasks. Finally, a union loss function is designed to enhance the localization ability. Experiments on the MS COCO dataset demonstrate that the proposed MRFENet has a significant improvement against other state-of-theart methods, which further validates that MRFENet can effectively utilize small object information.

SOCR‐YOLO: Small Objects Detection Algorithm in Medical Images

“…Focal FasterNet block (FFNB) achieves precise detection of small objects in drone‐captured scenes by merging shallow and deep features [38]. The Multilevel Receptive Field Expansion Network (MRFENet) developed by Liu et al employs Receptive Field Expansion Blocks (RFEB) to explore contextual clues and details on small objects [39]. Additionally, Wan et al introduced a novel approach for small object detection that relies on density‐aware scale adaptation, transitioning from a coarse‐to‐fine detection process [40].…”

Textile defect detection based on multi‐proportion spatial attention mechanism and channel memory feature fusion network

“…In [7], the authors combine defect size and the k-means algorithm to perform dimension clustering on target images and add YOLO detection layers to the feature maps for textile defect detection. In [8], the authors presented a multilayer receptive field expansion network to increase detection performance for tiny objects and introduced Swin Transformer as the backbone network to extract features.…”