Orientation robust object detection in aerial images using deep convolutional neural network

By Haigang Zhu, Xiaogang Chen, Weiqun Dai, Kun Fu, Qixiang Ye, Jianbin Jiao. International Conference on Information Photonics, 2015

386 citations recorded by the source corpus.

Subfields: Aerospace Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/ffc2f8c1bf568b913427d1702d75306311df3d5b

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

1. Xue Yang
2. Junchi Yan
3. Guisong Xia
4. Jian Ding
5. Wentao Wang
6. Qi Ming
7. Nan Xue
8. Qi Tian

In Computer Vision and Pattern Recognition

1. Junwei Han
2. Junchi Yan
3. Gong Cheng
4. Luc Van Gool
5. Xue Yang
6. Fahad Shahbaz Khan
7. Guisong Xia
8. Dacheng Tao

Citations in the corpus

• Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence (2021, background): “UCAS-AOD contains 1,510 aerial images of approximately 659× 1, 280 pixels, with two categories of 14,596 instances in total.”
• EarthGPT: A Universal Multimodal Large Language Model for Multisensor Image Comprehension in Remote Sensing Domain (2024, background): “For other tasks such as VQA, RSVG, and HBB detection, the two-step training strategy brings EarthGPT an increase of 3.82% (OA), 4.06% (mIoU), and 3.74% (AP@50) on RSVQA-HR, DIOR-RSVG, and MAR20 datasets, respectively.”
• Deep learning in remote sensing: a review (2017, methodology): “[104] have employed the pretrained CNN features that are extracted from combined layers and implemented orientation-robust object detection in a coarse localization framework.”
• Transformers in Remote Sensing: A Survey (2022, methodology): “Oriented RepPoints [46] Detection DOTA [49], UCAS-AOD [86], HRSC2016 [81] mAP Proposes an anchor-free detector learns flexible adaptive points as representations through a quality assessment and sample assignment scheme.”
• Rethinking Rotated Object Detection with Gaussian Wasserstein Distance Loss (2021, background): “UCAS-AOD [83] contains 1,510 aerial images of about 659 × 1, 280 pixels, with 2 categories of 14,596 instances.”
• SCRDet++: Detecting Small, Cluttered and Rotated Objects via Instance-Level Feature Denoising and Rotation Loss Smoothing (2020, methodology): “By combing these two features, our proposed detector is termed as SCRDet++. Extensive experiments are performed on large aerial images public datasets DOTA, DIOR, UCAS-AOD as well as natural image dataset COCO, scene text dataset ICDAR2015, small traffic light dataset BSTLD and our released S 2 TLD ”
• Object Detection in Optical Remote Sensing Images: A Survey and A New Benchmark (2019, methodology): “Although several popular object detection datasets, such as NWPU VHR ‐ 10 (Cheng et al., 2016a), UCAS ‐ AOD (Zhu et al., 2015a), COWC (Mundhenk et al., 2016), and DOTA (Xia et al., 2018), are proposed in the earth observation community, they are still far from satisfying the requirements of deep…”
• Towards Multi-class Object Detection in Unconstrained Remote Sensing Imagery (2018, methodology): “HEM) [31] and we use Soft NMS [32] only for the HBB benchmark. Towards Multi-class Object Detection in Unconstrained RS Imagery 9 3.1 Datasets The experiments were conducted on the DOTA [4], UCAS-AOD [33], and NWPU VHR-10 [34] image datasets which all have multi-class object annotations. DOTA is so ”
• Hierarchical and Robust Convolutional Neural Network for Very High-Resolution Remote Sensing Object Detection (2019, methodology): “To deal with this problem, an orientation robust feature extractor is proposed in [46], and the rotation-invariant CNNs (RICNNs) structure containing a rotation-invariant fully connected layer is introduced [2].”
• Accurate Object Localization in Remote Sensing Images Based on Convolutional Neural Networks (2017, methodology): “[43] used CNN features from combined layers to perform orientation-robust aerial object detection.”
• Learning Modulated Loss for Rotated Object Detection (2019, methodology): “d ship close inshore. The size of each image ranges from 300 300 to 1;500 900. This dataset has 1061 images including 436 images for training, 181 images for validation, and 444 for testing. UCAS-AOD [49]: UCAS-AOD is a remote sensing dataset which contains two categories: car and plane. UCAS-AOD co”
• Object Detection in Aerial Images: A Large-Scale Benchmark and Challenges (2021, background): “, [7], [15], [16], [17], [26], but they contain a limited number of instances and tend to use images taken under ideal conditions (e.”
• DOTA: A Large-Scale Dataset for Object Detection in Aerial Images (2017, methodology): “Annotation way #main categories #Instances #Images Image width NWPU VHR-10 [2] horizontal BB 10 3651 800 ∼ 1000 SZTAKI-INRIA [1] oriented BB 1 665 9 ∼ 800 TAS [9] horizontal BB 1 1319 30 792 COWC [20] one dot 1 32716 53 2000 ∼ 19,000 VEDAI [24] oriented BB 3 2950 1268 512, 1024 UCAS-AOD [40] oriente”
• FAIR1M: A Benchmark Dataset for Fine-grained Object Recognition in High-Resolution Remote Sensing Imagery (2021, methodology): “The UCAS-AOD dataset [67] is designed for air-plane and vehicle detection, which consists of an airplane dataset and a vehicle dataset.”
• R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object (2019, background): “UCAS-AOD [45] contains 1510 aerial images of approximately 659 × 1280 pixels, it contains two categories of 14596 instances.”
• Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (2020, background): “The total iterations of HRSC2016, DOTA, UCAS-AOD, and ICDAR 2015 are 20k, 30k, 15k, and 40k, respectively.”
• PIoU Loss: Towards Accurate Oriented Object Detection in Complex Environments (2020, methodology): “SZTAKI [1] Aerial ≈1:3 9 665 VEDAI [31] Aerial 1:3 1268 2950 UCAS-AOD [50] Aerial 1:1.”
• iSAID: A Large-scale Dataset for Instance Segmentation in Aerial Images (2019, methodology): “ bridge, large vehicle, small vehicle, helicopter, roundabout, swimming pool and soccer ball field. Objects from these categories occur frequently and are important for various real-world applications [5, 23, 35]. For dataset splits, we use half of the original images to form train set, 1=6 images fo”
• Oriented RepPoints for Aerial Object Detection (2021, methodology): “UCAS-AOD datasets contains a large number of small objects with complex surrounding scenes.”