EFN: Field-Based Object Detection for Aerial Images

Liu, Jin; Zheng, Haokun

doi:10.3390/rs12213630

Cited by 8 publications

(5 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Huang et al [19] developed a novel object-adaptation label assignment strategy that is flexible to fit the object's size and direction. EFN [58] employed an ellipse field network that integrates semantic segmentation and object detection. Yi et al [31] defined the BBAVectors to represent OOB.…”

Section: Single-stage Anchor-free Methods For Oodmentioning

confidence: 99%

Dual-Resolution and Deformable Multihead Network for Oriented Object Detection in Remote Sensing Images

Liu

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Compared with general object detection, the scale variations, arbitrary orientations, and complex backgrounds of objects in remote sensing images make it more challenging to detect oriented objects. Especially for oriented objects that have large aspect ratios, it is more difficult to accurately detect their boundary. Many methods show excellent performance on oriented object detection, most of which are anchor-based algorithms. To mitigate the performance gap between anchor-free algorithms and anchor-based algorithms, this paper proposes an anchor-free algorithm called dual-resolution and deformable multi-head network (DDMNet) for oriented object detection. Specifically, the dual-resolution network with bilateral fusion is adopted to extract high-resolution feature maps which contain both spatial details and multi-scale contextual information. Then, the deformable convolution is incorporated into the network to alleviate the misalignment problem of oriented object detection. And a dilated feature fusion module is performed on the deformable feature maps to expand their receptive fields. Finally, box boundary-aware vectors instead of the angle are leveraged to represent the oriented bounding box and the multi-head network is designed to get robust predictions. DDMNet is a single-stage oriented object detection method without using anchors and exhibits promising performance on the public challenging benchmarks. DDMNet obtains 90.49%, 93.25%, and 78.66% mean average precision (mAP) on the HRSC2016, FGSD2021, and DOTA datasets. In particular, DDMNet achieves 79.86% at mAP75 and 53.85% at mAP85 on the HRSC2016 dataset, respectively, outperforming the current state-of-the-art methods.

show abstract

Section: Single-stage Anchor-free Methods For Oodmentioning

confidence: 99%

Dual-Resolution and Deformable Multihead Network for Oriented Object Detection in Remote Sensing Images

Liu

et al. 2023

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

show abstract

“…Some methods discard the anchor boxes, e.g., semantic segmentation methods to extract the object contours and then transformed them into OBBs by post-processing [47][48][49]. EFN [50] applies the ellipse field to segment the target and calculates the rotation angle through the major axis and minor axis of the ellipse. By using the above methods, the object detection performance could be improved apparently.…”

Section: Oriented Object Detectionmentioning

confidence: 99%

HODet: A New Detector for Arbitrary-Oriented Rectangular Object in Optical Remote Sensing Imagery

Xu,

Yu,

Ming

2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Object detection from remote sensing images is a key technology for earth observation applications, which has important scientific research value. Ground objects in remote sensing images appear at arbitrary angles. However, object detection based on horizontal bounding boxes (HBBs) would cause mutual coverage among targets while the ground objects were dense-distributed or with a large aspect ratio. The oriented object detection methods could solve the problem by predicting the rotation angle. But the currently used methods were timeconsuming in labelling and require complex loss functions of networks. Thus, this paper combined HBB and oriented object detection to achieve rectangular object detection. Firstly, object detection network with HBBs is trained and the model predicts results in original remote sensing image. Secondly, the rotation angles are derived from the line detection with linear Hough transform on the cropped region of targets obtained by the first step. Then, the original image is rotated according to the rotation angles and detects object with HBBs again. Finally, new bounding boxes are mapped to the original image to get the detection results with oriented bounding boxes (OBBs).Experiments on public remote sensing datasets demonstrate that the proposed method is highly flexible and can be combined with any HBBs object detection network. The idea of original image rotation does not use the OBB labels to retrain the network, which reduces the workload of object annotations. Additionally, the method can be used to automatically generate OBBs to label rectangular objects since the public datasets commonly annotated by HBBs.

show abstract

“…[10][11][12]33] have focused on solving the discontinuous problem of angle parameter regression in the training process, and constantly refining the loss function to improve performance. Some other scholars attempt to use a target representation vector that can eliminate such boundary discontinuities to represent instances, such as polar coordinates [12], ellipse bounding box [34], and middle lines of boxes [35]. Additionally, to obtain better-refined rotated anchor boxes, RR-CNN [36], R3Det [37], and CFC-Net [38] focus on spatial alignment and anchor refinement to guide the training process.…”

Section: Related Work 21 Arbitrary Oriented Object Detectionmentioning

confidence: 99%

AFRE-Net: Adaptive Feature Representation Enhancement for Arbitrary Oriented Object Detection

Zhang,

Sun,

Zhuang

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Arbitrary-oriented object detection (AOOD) is a crucial task in aerial image analysis but is also faced with significant challenges. In current AOOD detectors, commonly used multi-scale feature fusion modules fall short in spatial and semantic information complement between scales. Additionally, fixed feature extraction structures are usually used following a fusion model, resulting in the inability of detectors to self-adjust. At the same time, feature fusion and extraction modules are designed in isolation and the internal synergy between them is ignored. The above problems result in feature representation deficiency, thus affecting the overall detection precision. To solve these problems, we first create a fine-grained feature pyramid network (FG-FPN) that not only provides richer spatial and semantic features, but also completes neighbor scale features in a self-learning mode. Subsequently, we propose a novel feature enhancement module (FEM) to fit FG-FPN. FEM authorizes the detection unit to automatically adjust the sensing area and adaptively suppress background interference, thereby generating stronger feature representations. Our proposed solution was tested through extensive experiments on challenging datasets, including DOTA (77.44% mAP), HRSC2016 (97.82% mAP), UCAS-AOD (91.34% mAP), as well as ICDAR2015 (86.27% F-score) and its effectiveness and high applicability are verified on all the above datasets.

show abstract

EFN: Field-Based Object Detection for Aerial Images

Cited by 8 publications

References 52 publications

Dual-Resolution and Deformable Multihead Network for Oriented Object Detection in Remote Sensing Images

Dual-Resolution and Deformable Multihead Network for Oriented Object Detection in Remote Sensing Images

HODet: A New Detector for Arbitrary-Oriented Rectangular Object in Optical Remote Sensing Imagery

AFRE-Net: Adaptive Feature Representation Enhancement for Arbitrary Oriented Object Detection

Contact Info

Product

Resources

About