Journal
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
Volume 60, Issue -, Pages -Publisher
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TGRS.2022.3177255
Keywords
Feature extraction; Detectors; Object detection; Training; Convolution; Proposals; Interference; Aerial images; anchor-free detector; class-aware context aggregation (CCA); oriented feature refinement (OFR); soft label assignment (SLA)
Categories
Funding
- National Natural Science Foundation of China [62036007, 62176195, 61976166]
- Special Project on Technological Innovation and Application Development [cstc2020jscx-dxwtB0032]
- Chongqing Excellent Scientist Project [cstc2021ycjh-bgzxm0339]
Ask authors/readers for more resources
In this article, an anchor-free detector with feature enhancement and soft label assignment (FSDet) is introduced for object detection in aerial images. It addresses the challenges of arbitrary-oriented objects and imbalanced foreground-background distribution. The proposed method achieves competitive performance by aligning features with oriented objects, suppressing background context, and assigning soft labels to optimize training.
Object detection in aerial images, different from general object detection, faces several challenges, such as arbitrary-oriented objects and extremely imbalanced foreground-background distribution. Although some recent proposed aerial object detection methods achieve promising results, they are mainly anchor-based detectors that rely heavily on predefined anchor boxes, and the final detection performance is sensitive to anchor-related hyperparameters. In contrast, in this article, we present an anchor-free detector with feature enhancement and soft label assignment (FSDet), which adopts a simpler design and achieves competitive performance. Specifically, to address the feature misalignment for detecting oriented objects, we propose an oriented feature refinement (OFR) module to align the features with oriented objects. To alleviate the background issue, we design a class-aware context aggregation module to integrate the intraclass context information and suppress the background context. Moreover, we propose a soft label assignment mechanism to measure the weight of training samples within the arbitrary-oriented objects, which can concentrate more on representative items with regard to their potential to detect oriented objects, achieving a more stable optimization during training. Extensive experiments on several datasets suggest that the proposed method is superior to the state-of-the-art methods and achieves a better tradeoff between speed and accuracy.
Authors
I am an author on this paper
Click your name to claim this paper and add it to your profile.
Reviews
Recommended
No Data Available