SAFDet: a semi-anchor-free detector for effective detection of oriented objects in aerial images.

Fang, Zhenyu; Ren, Jinchang; Sun, He; Marshall, Stephen; Han, Junwei; Zhao, Huimin

Abstract

An oriented bounding box (OBB) is preferable over a horizontal bounding box (HBB) in accurate object detection. Most of existing works utilize a two-stage detector for locating the HBB and OBB, respectively, which have suffered from the misaligned horizontal proposals and the interference from complex backgrounds. To tackle these issues, region of interest transformer and attention models were proposed, yet they are extremely computationally intensive. To this end, we propose a semi-anchor-free detector (SAFDet) for object detection in aerial images, where a rotation-anchor-free-branch (RAFB) is used to enhance the foreground features via precisely regressing the OBB. Meanwhile, a center-prediction-module (CPM) is introduced for enhancing object localization and suppressing the background noise. Both RAFB and CPM are deployed during training, avoiding increased computational cost of inference. By evaluating on DOTA and HRSC2016 datasets, the efficacy of our approach has been fully validated for a good balance between the accuracy and computational cost.

Citation

FANG, Z., REN, J., SUN, H., MARSHALL, S., HAN, J. and ZHAO, H. 2020. SAFDet: a semi-anchor-free detector for effective detection of oriented objects in aerial images. Remote sensing [online], 12(19), article 3225. Available from: https://doi.org/10.3390/rs12193225

• Journal Article Type: Article
• Acceptance Date: Sep 29, 2020
• Online Publication Date: Oct 3, 2020
• Publication Date: Oct 1, 2020
• Deposit Date: May 2, 2022
• Publicly Available Date: May 2, 2022
• Journal: Remote Sensing
• Electronic ISSN: 2072-4292
• Publisher: MDPI
• Peer Reviewed: Peer Reviewed
• Volume: 12
• Issue: 19
• Article Number: 3225
• DOI: https://doi.org/10.3390/rs12193225
• Keywords: Rotate region; Convolutional neural network; Anchor free; Aerial object detection
• Public URL: https://rgu-repository.worktribe.com/output/1085577