Noise-to-Convex: A Hierarchical Framework for SAR Oriented Object Detection via Scattering Keypoint Feature Fusion and Convex Contour Refinement
Abstract
:1. Introduction
- (1)
- This article introduces a novel noise-to-convex detection paradigm, which employs a hierarchical three-level framework that integrates SAR-specific mechanisms with deep learning methods, progressively obtaining a fine-grained convex contour description of the blurred object outline.
- (2)
- A strong-scattering-region generation (SSRG) module is developed at the bottom level. This module utilizes a diffusion model to learn and construct the spatial distribution of strong scattering regions, enabling the direct identification of approximate object regions.
- (3)
- A scattering-keypoint feature fusion (SKFF) module is introduced at the middle level. This module adaptively detects scattering keypoints and fuses their features, effectively enhancing the object feature and reducing false alarms in complex scenes.
- (4)
- A convex contour representation and convex contour prediction (CCP) module is proposed at the top level to obtain a precise and fine-grained convex contour for the object outline.
2. Related Work
2.1. Modeling Methods for SAR Object Outline
2.2. Context Modeling Methods for SAR Object Detection
2.3. Diffusion Models
3. Methods
3.1. Strong-Scattering-Region Generation Module
3.1.1. Strong-Scattering-Region Spatial Distribution Learning
3.1.2. Denoising Decoder
- (a)
- The embedding of the noise timestep t is concatenated with the input object queries and fed into the denoising decoder as new queries. This allows the module to implicitly learn the relationships between object features under varying noise conditions. Specifically, in the first layer, the time embedding is directly concatenated with the input object queries. In subsequent layers, the time embedding is concatenated with the processed object queries from the previous layer.
- (b)
- Inspired by the widespread usage of adaptive normalization layers [42] in GANs and diffusion models, we explore replacing all standard normalization layers in the denoising decoder with adaptive layer normalization (adaLN). Unlike standard normalization, adaLN conditions on the noise timestep t by regressing the scale and shift parameters and from its embedding. This approach applies the same conditioning function across all object queries, embedding the noise timestep information directly into the process of updating object features.
- (c)
- We replace only the standard normalization at the end of each denoising-decoder layer with adaLN, leaving other normalization operations unchanged. This introduces the noise timestep t solely at the layer outputs, reducing the number of parameters and computational complexity compared to replacing all standard normalization layers. We adopt this lightweight conditioning mechanism in our final design, as it achieves optimal performance according to our ablation studies.
3.2. Scattering-Keypoint Feature Fusion Module
3.2.1. Scattering-Keypoint Detection
3.2.2. Scattering-Keypoint Feature Fusion
3.3. Convex Contour Prediction Module
3.3.1. Convex Contour Representation
3.3.2. Convex Contour Regression
3.4. Dynamically Weighted CIoU Loss
4. Experiments and Analysis
4.1. Dataset
- (1)
- HRSID: The publicly available HRSID dataset is employed for SAR ship detection, semantic segmentation, and instance segmentation tasks. It consists of 5604 SAR images with resolution ranging from 0.5 to 3 m, and contains 16,951 labeled ship targets of varying sizes. The average size of the ship targets is approximately 33 * 29 pixels, and the proportions of small, medium, and large ship targets are 31.6%, 67.4%, and 1.0%, respectively.
- (2)
- RSDD-SAR: The publicly released RSDD-SAR dataset consists of 84 scenes of GF-3 data slices, 41 scenes of TerraSAR-X data slices and 2 scenes of large uncropped images, including 7000 slices and 10,263 ship instances of multi-observing modes, multi-polarization modes, and multi-resolutions. The average size of the ship targets is approximately 34 * 28 pixels, and the proportions of small, medium, and large ship targets are 22.7%, 77.2%, and 0.1%, respectively.
- (3)
- SAR-Aircraft-v1.0: The public SAR-Aircraft-v1.0 is a high-resolution SAR aircraft dataset. It consists of 4368 images obtained from the GF-3 satellite and 16,463 labeled aircraft instances, covering seven aircraft categories. In this article, in order to better verify the effectiveness of our proposed method, we simply merge them into one single plane category. The average size of the aircraft targets is approximately 81 * 85 pixels, and the proportions of small, medium, and large aircraft targets are 0.2%, 76.4%, and 23.4%, respectively.
4.2. Experimental Setup and Parameters
4.3. Evaluation Metrics
4.4. Comparison Experiments
4.4.1. Comparison Results on HRSID
4.4.2. Comparison Results on RSDD-SAR
4.4.3. Comparison Results on SAR-Aircraft-v1.0
4.5. Analysis of Visualization Results
4.6. Ablation Studies
4.6.1. CCP Module Ablation Experiment
4.6.2. SSRG Module Ablation Experiment
4.6.3. SKFF Module Ablation Experiment
4.6.4. Number of Queries Ablation Experiment
5. Discussion
- Robustness to Noise. SAR imagery in maritime surveillance or disaster monitoring typically suffers from significant speckle noise and clutter, but our hierarchical approach effectively filters out irrelevant background regions to produce accurate detections with reduced false alarms.
- Reduced False Alarms. By leveraging geometric priors through adaptive scattering-keypoint detection and convex contour modeling, our method reduces the number of spurious detections, which is crucial for applications such as ship detection in congested sea lanes or vehicle recognition in complex urban environments.
- Adaptability to Varying Annotation Styles. Although our primary contribution focuses on oriented bounding boxes, the proposed method can be adapted to horizontal bounding boxes or other annotation formats with minimal changes, as demonstrated by experiments on the SAR-Aircraft-v1.0 dataset.
- Enhanced Tiny Object Detection. We plan to enhance the hierarchical design with multi-scale feature aggregation or context-aware attention strategies to better capture faint scattering signatures of very small objects.
- Model Optimization. We will investigate techniques such as model compression, knowledge distillation, and more efficient attention variants to facilitate easier deployment while maintaining detection accuracy. These optimizations will enhance the feasibility of our approach for on-board SAR processing and real-time surveillance.
- Exploration of Diffusion Models. Our experiments suggest that diffusion-based methods hold promise for enhancing both generative and discriminative capabilities in noisy SAR imagery. We plan to deeply integrate diffusion models with the underlying mechanisms of SAR imagery to further improve performance under severe clutter.
6. Conclusions
Author Contributions
Funding
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Zhou, X.; Li, T. Ship detection in PolSAR images based on a modified polarimetric notch filter. Electronics 2023, 12, 2683. [Google Scholar] [CrossRef]
- Min-Woo, K.; Park, S.K.; Ju, J.G.; Noh, H.C.; Choi, D.G. Clean Collector Algorithm for Satellite Image Pre-Processing of SAR-to-EO Translation. Electronics 2024, 13, 4529. [Google Scholar] [CrossRef]
- Yang, C.; Wang, D.; Sun, F.; Wang, K. Maneuvering Trajectory Synthetic Aperture Radar Processing Based on the Decomposition of Transfer Functions in the Frequency Domain Using Average Blurred Edge Width Assessment. Electronics 2024, 13, 4100. [Google Scholar] [CrossRef]
- Sun, S.; Wang, J. Ship Detection in SAR Images Based on Steady CFAR Detector and Knowledge-Oriented GBDT Classifier. Electronics 2024, 13, 2692. [Google Scholar] [CrossRef]
- Zheng, H.; Xue, X.; Yue, R.; Liu, C.; Liu, Z. SAR Image Ship Target Detection Based on Receptive Field Enhancement Module and Cross-Layer Feature Fusion. Electronics 2023, 13, 167. [Google Scholar] [CrossRef]
- Gao, G.; Liu, L.; Zhao, L.; Shi, G.; Kuang, G. An adaptive and fast CFAR algorithm based on automatic censoring for target detection in high-resolution SAR images. IEEE Trans. Geosci. Remote Sens. 2008, 47, 1685–1697. [Google Scholar] [CrossRef]
- Li, Q.; Mou, L.; Liu, Q.; Wang, Y.; Zhu, X.X. HSF-Net: Multiscale deep feature embedding for ship detection in optical remote sensing imagery. IEEE Trans. Geosci. Remote Sens. 2018, 56, 7147–7161. [Google Scholar] [CrossRef]
- Nieto-Hidalgo, M.; Gallego, A.J.; Gil, P.; Pertusa, A. Two-stage convolutional neural network for ship and spill detection using SLAR images. IEEE Trans. Geosci. Remote Sens. 2018, 56, 5217–5230. [Google Scholar] [CrossRef]
- Du, L.; Li, L.; Wei, D.; Mao, J. Saliency-guided single shot multibox detector for target detection in SAR images. IEEE Trans. Geosci. Remote Sens. 2019, 58, 3366–3376. [Google Scholar] [CrossRef]
- Ai, J.; Tian, R.; Luo, Q.; Jin, J.; Tang, B. Multi-scale rotation-invariant Haar-like feature integrated CNN-based ship detection algorithm of multiple-target environment in SAR imagery. IEEE Trans. Geosci. Remote Sens. 2019, 57, 10070–10087. [Google Scholar] [CrossRef]
- Ren, S.; He, K.; Girshick, R.; Sun, J. Faster r-cnn: Towards real-time object detection with region proposal networks. IEEE Trans. Pattern Anal. Mach. Intell. 2017, 39, 1137–1149. [Google Scholar] [CrossRef]
- Redmon, J.; Divvala, S.; Girshick, R.; Farhadi, A. You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; pp. 779–788. [Google Scholar]
- Carion, N.; Massa, F.; Synnaeve, G.; Usunier, N.; Kirillov, A.; Zagoruyko, S. End-to-end object detection with transformers. In Proceedings of the European Conference on Computer Vision, Glasgow, UK, 23–28 August 2020; Springer: Berlin/Heidelberg, Germany, 2020; pp. 213–229. [Google Scholar]
- Wang, Z.; Du, L.; Mao, J.; Liu, B.; Yang, D. SAR target detection based on SSD with data augmentation and transfer learning. IEEE Geosci. Remote Sens. Lett. 2018, 16, 150–154. [Google Scholar] [CrossRef]
- Li, J.; Qu, C.; Shao, J. Ship detection in SAR images based on an improved faster R-CNN. In Proceedings of the IEEE 2017 SAR in Big Data Era: Models, Methods and Applications (BIGSARDATA), Beijing, China, 13–14 November 2017; pp. 1–6. [Google Scholar]
- Jiao, J.; Zhang, Y.; Sun, H.; Yang, X.; Gao, X.; Hong, W.; Fu, K.; Sun, X. A densely connected end-to-end neural network for multiscale and multiscene SAR ship detection. IEEE Access 2018, 6, 20881–20892. [Google Scholar] [CrossRef]
- An, Q.; Pan, Z.; Liu, L.; You, H. DRBox-v2: An improved detector with rotatable boxes for target detection in SAR images. IEEE Trans. Geosci. Remote Sens. 2019, 57, 8333–8349. [Google Scholar] [CrossRef]
- Sun, Y.; Wang, Z.; Sun, X.; Fu, K. SPAN: Strong scattering point aware network for ship detection and classification in large-scale SAR imagery. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2022, 15, 1188–1204. [Google Scholar] [CrossRef]
- Zhang, C.; Yang, C.; Cheng, K.; Guan, N.; Dong, H.; Deng, B. MSIF: Multisize inference fusion-based false alarm elimination for ship detection in large-scale SAR images. IEEE Trans. Geosci. Remote Sens. 2022, 60, 1–11. [Google Scholar] [CrossRef]
- Cui, Z.; Li, Q.; Cao, Z.; Liu, N. Dense attention pyramid networks for multi-scale ship detection in SAR images. IEEE Trans. Geosci. Remote Sens. 2019, 57, 8983–8997. [Google Scholar] [CrossRef]
- Kang, Y.; Wang, Z.; Fu, J.; Sun, X.; Fu, K. SFR-Net: Scattering feature relation network for aircraft detection in complex SAR images. IEEE Trans. Geosci. Remote Sens. 2021, 60, 1–17. [Google Scholar] [CrossRef]
- Yang, X.; Zhang, X.; Wang, N.; Gao, X. A robust one-stage detector for multiscale ship detection with complex background in massive SAR images. IEEE Trans. Geosci. Remote Sens. 2021, 60, 1–12. [Google Scholar] [CrossRef]
- Sun, Y.; Sun, X.; Wang, Z.; Fu, K. Oriented ship detection based on strong scattering points network in large-scale SAR images. IEEE Trans. Geosci. Remote Sens. 2021, 60, 1–18. [Google Scholar] [CrossRef]
- Guo, P.; Celik, T.; Liu, N.; Li, H.C. Break through the border restriction of horizontal bounding box for arbitrary-oriented ship detection in SAR images. IEEE Geosci. Remote Sens. Lett. 2023, 20, 1–5. [Google Scholar] [CrossRef]
- Zhao, Y.; Zhao, L.; Liu, Z.; Hu, D.; Kuang, G.; Liu, L. Attentional feature refinement and alignment network for aircraft detection in SAR imagery. arXiv 2022, arXiv:2201.07124. [Google Scholar] [CrossRef]
- Fu, K.; Fu, J.; Wang, Z.; Sun, X. Scattering-keypoint-guided network for oriented ship detection in high-resolution and large-scale SAR images. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 11162–11178. [Google Scholar] [CrossRef]
- Pan, D.; Gao, X.; Dai, W.; Fu, J.; Wang, Z.; Sun, X.; Wu, Y. SRT-Net: Scattering region topology network for oriented ship detection in large-scale SAR images. IEEE Trans. Geosci. Remote. Sens. 2024, 62, 5202318. [Google Scholar] [CrossRef]
- Ju, M.; Niu, B.; Zhang, J. FPDDet: An Efficient Rotated SAR Ship Detector Based on Simple Polar Encoding and Decoding. IEEE Trans. Geosci. Remote. Sens. 2023, 61, 5218915. [Google Scholar] [CrossRef]
- Wan, H.; Chen, J.; Huang, Z.; Du, W.; Xu, F.; Wang, F.; Wu, B. Orientation Detector for Small Ship Targets in SAR Images Based on Semantic Flow Feature Alignment and Gaussian Label Matching. IEEE Trans. Geosci. Remote. Sens. 2023, 61, 5218616. [Google Scholar] [CrossRef]
- Kang, M.; Ji, K.; Leng, X.; Lin, Z. Contextual region-based convolutional neural network with multilayer fusion for SAR ship detection. Remote Sens. 2017, 9, 860. [Google Scholar] [CrossRef]
- Chen, C.; He, C.; Hu, C.; Pei, H.; Jiao, L. MSARN: A deep neural network based on an adaptive recalibration mechanism for multiscale and arbitrary-oriented SAR ship detection. IEEE Access 2019, 7, 159262–159283. [Google Scholar] [CrossRef]
- Fu, K.; Dou, F.Z.; Li, H.C.; Diao, W.H.; Sun, X.; Xu, G.L. Aircraft recognition in SAR images based on scattering structure feature and template matching. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2018, 11, 4206–4217. [Google Scholar] [CrossRef]
- Ho, J.; Jain, A.; Abbeel, P. Denoising diffusion probabilistic models. Adv. Neural Inf. Process. Syst. 2020, 33, 6840–6851. [Google Scholar]
- Song, J.; Meng, C.; Ermon, S. Denoising diffusion implicit models. arXiv 2020, arXiv:2010.02502. [Google Scholar]
- Takagi, Y.; Nishimoto, S. High-resolution image reconstruction with latent diffusion models from human brain activity. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Vancouver, BC, Canada, 18–22 June 2023; pp. 14453–14463. [Google Scholar]
- Gu, Z.; Chen, H.; Xu, Z. Diffusioninst: Diffusion model for instance segmentation. In Proceedings of the ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Seoul, Republic of Korea, 14–19 April 2024; pp. 2730–2734. [Google Scholar]
- Chen, S.; Sun, P.; Song, Y.; Luo, P. Diffusiondet: Diffusion model for object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Paris, France, 2–6 October 2023; pp. 19830–19843. [Google Scholar]
- Perera, M.V.; Nair, N.G.; Bandara, W.G.C.; Patel, V.M. SAR despeckling using a denoising diffusion probabilistic model. IEEE Geosci. Remote Sens. Lett. 2023, 20, 1–5. [Google Scholar] [CrossRef]
- Zhang, X.; Li, Y.; Li, F.; Jiang, H.; Wang, Y.; Zhang, L.; Zheng, L.; Ding, Z. Ship-Go: SAR ship images inpainting via instance-to-image generative diffusion models. ISPRS J. Photogramm. Remote Sens. 2024, 207, 203–217. [Google Scholar] [CrossRef]
- Zhu, X.; Su, W.; Lu, L.; Li, B.; Wang, X.; Dai, J. Deformable detr: Deformable transformers for end-to-end object detection. arXiv 2020, arXiv:2010.04159. [Google Scholar]
- Song, Y.; Ermon, S. Improved techniques for training score-based generative models. Adv. Neural Inf. Process. Syst. 2020, 33, 12438–12448. [Google Scholar]
- Perez, E.; Strub, F.; De Vries, H.; Dumoulin, V.; Courville, A. Film: Visual reasoning with a general conditioning layer. In Proceedings of the AAAI Conference on Artificial Intelligence, New Orleans, LA, USA, 2–7 February 2018; Volume 32, pp. 3942–3951. [Google Scholar]
- Guo, Q.; Wang, H.; Xu, F. Scattering enhanced attention pyramid network for aircraft detection in SAR images. IEEE Trans. Geosci. Remote Sens. 2020, 59, 7570–7587. [Google Scholar] [CrossRef]
- Guo, Z.; Liu, C.; Zhang, X.; Jiao, J.; Ji, X.; Ye, Q. Beyond bounding-box: Convex-hull feature adaptation for oriented and densely packed object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Nashville, TN, USA, 20–25 June 2021; pp. 8792–8801. [Google Scholar]
- Jarvis, R.A. On the identification of the convex hull of a finite set of points in the plane. Inf. Process. Lett. 1973, 2, 18–21. [Google Scholar] [CrossRef]
- Lin, T.Y.; Goyal, P.; Girshick, R.; He, K.; Dollár, P. Focal loss for dense object detection. In Proceedings of the IEEE International Conference on Computer Vision, Venice, Italy, 22–29 October 2017; pp. 2980–2988. [Google Scholar]
- Wei, S.; Zeng, X.; Qu, Q.; Wang, M.; Su, H.; Shi, J. HRSID: A high-resolution SAR images dataset for ship detection and instance segmentation. IEEE Access 2020, 8, 120234–120254. [Google Scholar] [CrossRef]
- Congan, X.; Hang, S.; Jianwei, L.; Yu, L.; Libo, Y.; Long, G.; Wenjun, Y.; Taoyang, W. RSDD-SAR: Rotated ship detection dataset in SAR images. J. Radars 2022, 11, 581–599. [Google Scholar]
- Zhirui, W.; Yuzhuo, K.; Xuan, Z.; Yuelei, W.; Ting, Z.; Xian, S. SAR-AIRcraft-1.0: High-resolution SAR aircraft detection and recognition dataset. J. Radars 2023, 12, 906–922. [Google Scholar]
- Zhou, Y.; Yang, X.; Zhang, G.; Wang, J.; Liu, Y.; Hou, L.; Jiang, X.; Liu, X.; Yan, J.; Lyu, C.; et al. Mmrotate: A rotated object detection benchmark using pytorch. In Proceedings of the 30th ACM International Conference on Multimedia, Lisbon, Portugal, 10–14 October 2022; pp. 7331–7334. [Google Scholar]
- He, K.; Zhang, X.; Ren, S.; Sun, J. Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; pp. 770–778. [Google Scholar]
- Deng, J.; Dong, W.; Socher, R.; Li, L.J.; Li, K.; Fei-Fei, L. Imagenet: A large-scale hierarchical image database. In Proceedings of the 2009 IEEE Conference on Computer Vision and Pattern Recognition, Miami, FL, USA, 20–25 June 2009; pp. 248–255. [Google Scholar]
- Lin, T.Y.; Maire, M.; Belongie, S.; Hays, J.; Perona, P.; Ramanan, D.; Dollár, P.; Zitnick, C.L. Microsoft coco: Common objects in context. In Proceedings of the Computer Vision–ECCV 2014: 13th European Conference, Zurich, Switzerland, 6–12 September 2014; Proceedings, Part V 13. Springer: Berlin/Heidelberg, Germany, 2014; pp. 740–755. [Google Scholar]
- Li, W.; Chen, Y.; Hu, K.; Zhu, J. Oriented reppoints for aerial object detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, New Orleans, LA, USA, 18–24 June 2022; pp. 1829–1838. [Google Scholar]
- Xie, X.; Cheng, G.; Wang, J.; Yao, X.; Han, J. Oriented R-CNN for object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Montreal, BC, Canada, 11–17 October 2021; pp. 3520–3529. [Google Scholar]
- Zhu, J.; Jing, D.; Gao, D. Stage-by-Stage Adaptive Alignment Mechanism for Object Detection in Aerial Images. Electronics 2024, 13, 3640. [Google Scholar] [CrossRef]
- Xia, G.S.; Bai, X.; Ding, J.; Zhu, Z.; Belongie, S.; Luo, J.; Datcu, M.; Pelillo, M.; Zhang, L. DOTA: A large-scale dataset for object detection in aerial images. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–22 June 2018; pp. 3974–3983. [Google Scholar]
- Dai, L.; Liu, H.; Tang, H.; Wu, Z.; Song, P. Ao2-detr: Arbitrary-oriented object detection transformer. IEEE Trans. Circuits Syst. Video Technol. 2022, 33, 2342–2356. [Google Scholar] [CrossRef]
- Tian, Z.; Shen, C.; Chen, H.; He, T. Fcos: Fully convolutional one-stage object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Republic of Korea, 27 October–2 November 2019; pp. 9627–9636. [Google Scholar]
- Xu, Y.; Fu, M.; Wang, Q.; Wang, Y.; Chen, K.; Xia, G.S.; Bai, X. Gliding vertex on the horizontal bounding box for multi-oriented object detection. IEEE Trans. Pattern Anal. Mach. Intell. 2020, 43, 1452–1459. [Google Scholar] [CrossRef] [PubMed]
- Feng, Y.; You, Y.; Tian, J.; Meng, G. OEGR-DETR: A Novel Detection Transformer Based on Orientation Enhancement and Group Relations for SAR Object Detection. Remote Sens. 2023, 16, 106. [Google Scholar] [CrossRef]
- Zhang, J.; Xing, M.; Sun, G.C.; Li, N. Oriented Gaussian function-based box boundary-aware vectors for oriented ship detection in multiresolution SAR imagery. IEEE Trans. Geosci. Remote Sens. 2021, 60, 1–15. [Google Scholar] [CrossRef]
- Cai, Z.; Vasconcelos, N. Cascade r-cnn: Delving into high quality object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, USA, 18–22 June 2018; pp. 6154–6162. [Google Scholar]
- Yang, Z.; Liu, S.; Hu, H.; Wang, L.; Lin, S. Reppoints: Point set representation for object detection. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Seoul, Republic of Korea, 27 October–2 November 2019; pp. 9657–9666. [Google Scholar]
- Zhang, C.Q.; Deng, Y.; Chong, M.Z.; Zhang, Z.W.; Tan, Y.H. Entropy-Based re-sampling method on SAR class imbalance target detection. ISPRS J. Photogramm. Remote Sens. 2024, 209, 432–447. [Google Scholar] [CrossRef]
Method | HRSID | RSDD-SAR | ||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Precision | Recall | F1 | Precision | Recall | F1 | |||||||||||
Optical OBB Detectors: | ||||||||||||||||
Faster RCNN-O [57] | 0.604 | 0.828 | 0.698 | 0.782 | 0.342 | 0.375 | 0.46 | 0.057 | 0.752 | 0.907 | 0.822 | 0.879 | 0.424 | 0.434 | 0.539 | 0.252 |
RoI-Transformer [14] | 0.644 | 0.83 | 0.725 | 0.809 | 0.435 | 0.412 | 0.551 | 0.077 | 0.817 | 0.922 | 0.866 | 0.907 | 0.507 | 0.469 | 0.603 | 0.502 |
Deformable DETR-O [58] | 0.212 | 0.889 | 0.342 | 0.735 | 0.305 | 0.345 | 0.435 | 0.023 | 0.316 | 0.929 | 0.472 | 0.835 | 0.357 | 0.402 | 0.480 | 0.375 |
FCOS-O [59] | 0.393 | 0.827 | 0.533 | 0.784 | 0.378 | 0.37 | 0.51 | 0.12 | 0.630 | 0.909 | 0.744 | 0.882 | 0.452 | 0.436 | 0.581 | 0.353 |
RetinaNet-O [46] | 0.217 | 0.811 | 0.342 | 0.745 | 0.298 | 0.342 | 0.431 | 0.11 | 0.617 | 0.898 | 0.731 | 0.852 | 0.374 | 0.412 | 0.476 | 0.404 |
ReDet [15] | 0.733 | 0.865 | 0.794 | 0.844 | 0.513 | 0.469 | 0.591 | 0.197 | 0.804 | 0.935 | 0.865 | 0.920 | 0.544 | 0.487 | 0.642 | 0.454 |
CFA [44] | 0.275 | 0.889 | 0.420 | 0.846 | 0.404 | 0.413 | 0.549 | 0.218 | 0.448 | 0.897 | 0.598 | 0.866 | 0.356 | 0.390 | 0.562 | 0.950 |
Oriented RepPoints [54] | 0.336 | 0.899 | 0.489 | 0.858 | 0.441 | 0.424 | 0.56 | 0.301 | 0.315 | 0.898 | 0.466 | 0.869 | 0.386 | 0.410 | 0.551 | 0.850 |
Oriented RCNN [55] | 0.644 | 0.830 | 0.725 | 0.809 | 0.435 | 0.412 | 0.551 | 0.077 | 0.833 | 0.930 | 0.879 | 0.890 | 0.510 | 0.469 | 0.606 | 0.403 |
Gliding Vertex [60] | 0.545 | 0.823 | 0.656 | 0.755 | 0.333 | 0.355 | 0.475 | 0.072 | 0.594 | 0.925 | 0.723 | 0.847 | 0.424 | 0.425 | 0.546 | 0.338 |
SSADet [56] | 0.702 | 0.869 | 0.777 | 0.842 | 0.461 | 0.435 | 0.571 | 0.255 | 0.680 | 0.940 | 0.789 | 0.917 | 0.518 | 0.481 | 0.622 | 0.750 |
SAR OBB Detectors: | ||||||||||||||||
OG-BBAV [62] | 0.257 | 0.935 | 0.403 | 0.798 | 0.388 | 0.392 | 0.494 | 0.17 | 0.587 | 0.939 | 0.722 | 0.876 | 0.461 | 0.438 | 0.594 | 0.651 |
DCMSNN [16] | 0.229 | 0.933 | 0.368 | 0.796 | 0.429 | 0.416 | 0.506 | 0.087 | 0.522 | 0.939 | 0.671 | 0.881 | 0.452 | 0.440 | 0.552 | 0.353 |
Drbox v2 [17] | 0.2 | 0.885 | 0.326 | 0.678 | 0.257 | 0.322 | 0.349 | 0.001 | 0.253 | 0.880 | 0.393 | 0.710 | 0.146 | 0.291 | 0.259 | 0.002 |
FPDDet [28] | 0.745 | 0.848 | 0.793 | 0.826 | 0.479 | 0.432 | 0.583 | 0.244 | 0.847 | 0.933 | 0.888 | 0.921 | 0.542 | 0.481 | 0.642 | 0.352 |
FADet [29] | 0.715 | 0.877 | 0.788 | 0.853 | 0.482 | 0.492 | 0.573 | 0.323 | 0.848 | 0.925 | 0.885 | 0.909 | 0.479 | 0.459 | 0.601 | 0.801 |
OEGR-DETR [61] | 0.627 | 0.877 | 0.731 | 0.851 | 0.488 | 0.433 | 0.569 | 0.302 | 0.703 | 0.931 | 0.801 | 0.911 | 0.502 | 0.466 | 0.617 | 0.701 |
Ours: | ||||||||||||||||
SKG-DDT | 0.808 | 0.881 | 0.843 | 0.865 | 0.515 | 0.467 | 0.584 | 0.356 | 0.862 | 0.921 | 0.891 | 0.927 | 0.463 | 0.461 | 0.643 | 0.950 |
Method | Precision | Recall | F1 | |||||
---|---|---|---|---|---|---|---|---|
Faster RCNN [11] | 0.88 | 0.843 | 0.861 | 0.828 | 0.603 | 0.828 | 0.540 | 0.565 |
Cascade RCNN [63] | 0.902 | 0.831 | 0.865 | 0.817 | 0.639 | 0.813 | 0.552 | 0.568 |
Deformable DETR [40] | 0.801 | 0.905 | 0.850 | 0.878 | 0.531 | 0.255 | 0.528 | 0.462 |
FCOS [59] | 0.785 | 0.915 | 0.845 | 0.880 | 0.613 | 0.865 | 0.578 | 0.622 |
RetinaNet [46] | 0.720 | 0.880 | 0.792 | 0.850 | 0.603 | 0.842 | 0.555 | 0.588 |
RepPoints [64] | 0.754 | 0.897 | 0.819 | 0.870 | 0.641 | 0.790 | 0.569 | 0.598 |
SFR-Net [21] | 0.901 | 0.857 | 0.878 | 0.881 | 0.647 | 0.874 | 0.581 | 0.613 |
EBDet [65] | 0.894 | 0.896 | 0.895 | 0.882 | 0.651 | 0.901 | 0.580 | 0.644 |
Ours: | ||||||||
SKG-DDT | 0.912 | 0.852 | 0.881 | 0.892 | 0.637 | 0.514 | 0.591 | 0.659 |
CCP Module | SSRG Module | SKFF Module | Precision | Recall | F1 | |||||
---|---|---|---|---|---|---|---|---|---|---|
✘ | ✘ | ✘ | 0.076 | 0.909 | 0.140 | 0.735 | 0.305 | 0.345 | 0.435 | 0.023 |
✔ | ✘ | ✘ | 0.110 | 0.846 | 0.195 | 0.759 | 0.182 | 0.287 | 0.421 | 0.038 |
✔ | ✘ | ✔ | 0.127 | 0.862 | 0.221 | 0.813 | 0.221 | 0.321 | 0.448 | 0.088 |
✔ | ✔ | ✘ | 0.684 | 0.845 | 0.745 | 0.805 | 0.459 | 0.406 | 0.570 | 0.216 |
✔ | ✔ | ✔ | 0.808 | 0.881 | 0.843 | 0.865 | 0.515 | 0.467 | 0.584 | 0.356 |
Precision | Recall | F1 | ||||||
---|---|---|---|---|---|---|---|---|
1.0 | 0.627 | 0.805 | 0.705 | 0.821 | 0.446 | 0.402 | 0.572 | 0.266 |
1.2 | 0.586 | 0.811 | 0.680 | 0.837 | 0.444 | 0.397 | 0.578 | 0.281 |
1.5 | 0.808 | 0.881 | 0.843 | 0.865 | 0.515 | 0.467 | 0.584 | 0.356 |
1.8 | 0.702 | 0.825 | 0.758 | 0.841 | 0.470 | 0.418 | 0.584 | 0.304 |
2.0 | 0.689 | 0.785 | 0.734 | 0.811 | 0.204 | 0.310 | 0.438 | 0.220 |
Precision | Recall | F1 | ||||||
---|---|---|---|---|---|---|---|---|
(a) | 0.675 | 0.832 | 0.745 | 0.833 | 0.473 | 0.420 | 0.586 | 0.248 |
(b) | 0.703 | 0.837 | 0.764 | 0.854 | 0.491 | 0.445 | 0.590 | 0.298 |
(c) | 0.808 | 0.881 | 0.843 | 0.865 | 0.515 | 0.467 | 0.584 | 0.356 |
K | Precision | Recall | F1 | |||||
---|---|---|---|---|---|---|---|---|
4 | 0.678 | 0.811 | 0.739 | 0.853 | 0.448 | 0.403 | 0.575 | 0.262 |
9 | 0.808 | 0.881 | 0.843 | 0.865 | 0.515 | 0.467 | 0.584 | 0.356 |
15 | 0.664 | 0.828 | 0.737 | 0.862 | 0.470 | 0.419 | 0.580 | 0.301 |
20 | 0.739 | 0.821 | 0.778 | 0.845 | 0.449 | 0.411 | 0.581 | 0.273 |
25 | 0.706 | 0.820 | 0.759 | 0.832 | 0.455 | 0.409 | 0.583 | 0.284 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Liu, S.; Tong, M.; He, B.; Jiang, J.; He, C. Noise-to-Convex: A Hierarchical Framework for SAR Oriented Object Detection via Scattering Keypoint Feature Fusion and Convex Contour Refinement. Electronics 2025, 14, 569. https://doi.org/10.3390/electronics14030569
Liu S, Tong M, He B, Jiang J, He C. Noise-to-Convex: A Hierarchical Framework for SAR Oriented Object Detection via Scattering Keypoint Feature Fusion and Convex Contour Refinement. Electronics. 2025; 14(3):569. https://doi.org/10.3390/electronics14030569
Chicago/Turabian StyleLiu, Shuoyang, Ming Tong, Bokun He, Jiu Jiang, and Chu He. 2025. "Noise-to-Convex: A Hierarchical Framework for SAR Oriented Object Detection via Scattering Keypoint Feature Fusion and Convex Contour Refinement" Electronics 14, no. 3: 569. https://doi.org/10.3390/electronics14030569
APA StyleLiu, S., Tong, M., He, B., Jiang, J., & He, C. (2025). Noise-to-Convex: A Hierarchical Framework for SAR Oriented Object Detection via Scattering Keypoint Feature Fusion and Convex Contour Refinement. Electronics, 14(3), 569. https://doi.org/10.3390/electronics14030569