Article
Geochemistry & Geophysics
Zhiqing Li, Erzhu Li, Tianyu Xu, Alim Samat, Wei Liu
Summary: In this paper, a flow-guided upsampling module and a multifeature attention module are proposed to improve the feature pyramid network (FPN). The flow-guided upsampling module aligns features using flow warp, resulting in better cross-scale fusion. The multifeature attention module optimizes the weight of each level of features to reduce feature misalignment. Experimental results show that the improved FPN achieves better detection accuracy on three state-of-the-art models.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Environmental Sciences
Tianwen Zhang, Xiaoling Zhang, Xiao Ke
Summary: A novel quad feature pyramid network (Quad-FPN) is proposed for ship detection from synthetic aperture radar (SAR) imagery, with extensive ablation studies conducted to confirm its effectiveness. Experiments on five datasets show Quad-FPN's optimal performance compared to other 12 competitive CNN-based SAR ship detectors. Additionally, satisfactory detection results in actual ship detection further demonstrate Quad-FPN's practical application value in marine surveillance.
Article
Engineering, Electrical & Electronic
Runxi Wei, Zhejun Feng, Zengyan Wu, Chaoran Yu, Baoming Song, Changqing Cao
Summary: This article proposes an improved feature pyramid model called feature enhancement feature pyramid network (FE-FPN), which reduces information loss during the generation of feature maps and improves its capability to represent feature pyramids. The FE-FPN model achieves a 2.0% higher average precision compared to the feature pyramid network in Cascade R-CNN when using ResNet50. Quantitative comparisons with several classical characteristic pyramid models demonstrate the superior performance of the FE-FPN model.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Shuai Liu, Pengfei Chen, Yudong Zhang
Summary: Synthetic aperture radar (SAR) ship detection is widely used in various applications, but existing lightweight algorithms have problems such as misjudgment of targets mixed with the background and poor detection performance for targets with few samples. To address these issues, this paper proposes a detection network based on a multiscale feature pyramid network (FPN) that enhances target features, suppresses interference from the background, and reduces misjudgment. Experimental results show that the proposed network outperforms existing algorithms on public datasets.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Longbao Wang, Dongyu Gao, Xin Li, Xiaodan Tang, Xinyi Zhao, Hongmin Gao
Summary: Change detection is an essential task in remote sensing image processing, and fully convolutional networks are commonly used. However, downsampling in these networks leads to loss of spatial location information, impacting edge accuracy. To address computational and parameter inefficiencies, a lightweight composite convolution network (LWCCNet) is proposed, which incorporates composite convolution operators and a novel spatial attention mechanism for improved accuracy.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Qi Hu, Shaohai Hu, Shuaiqi Liu
Summary: In this paper, an anchor-free framework based on a balance attention network is proposed for multiscale ship detection in SAR images. The method utilizes deformable convolution and local attention module to capture local information of ships, and introduces a nonlocal attention module to extract nonlocal features of the SAR image. The feature pyramid network is used to detect ships of different sizes at different scales.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Haoming He, Yerui Chen, Mingchao Li, Qiang Chen
Summary: In this article, the authors propose ForkNet, an effective siamese feature pyramid network for remote sensing change detection. By applying FPN and CRAM in the feature extraction stage, ForkNet is able to generate feature representations with strong semantics and improve network performance. Additionally, the authors introduce pyramid Tversky loss, combined with Focal loss, to better train ForkNet and achieve state-of-the-art detection performance on challenging datasets.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Engineering, Multidisciplinary
Linyan Xue, Weina Yan, Ping Luo, Xiongfeng Zhang, Tetiana Chaikovska, Kun Liu, Wenshan Gao, Kun Yang
Summary: X-ray imaging is crucial for clinical diagnosis of hand fractures, as missed diagnosis may lead to severe consequences. A new guided anchoring method is proposed in this study, which significantly improves the detection performance of hand fractures.
ALEXANDRIA ENGINEERING JOURNAL
(2021)
Article
Forestry
Lin Zhang, Mingyang Wang, Yunhong Ding, Xiangfeng Bu
Summary: This paper proposes a multi-scale feature extraction model (MS-FRCNN) for small target forest fire detection by improving the classic Faster RCNN target detection model. The MS-FRCNN model utilizes ResNet50 as the backbone network to extract features and uses a new attention module (PAM) to reduce the influence of complex backgrounds and focus on the semantic and location information of small target forest fires. Experimental results show that the MS-FRCNN achieves better detection performance with a 5.7% higher accuracy compared to the baseline model.
Article
Geochemistry & Geophysics
Zhigang Yang, Junyu Kong, Binxi Zheng, Ming Li, Wei Emma Zhang, Tao Chen
Summary: This study proposes a region-based convolutional neural network RH-RCNN for arbitrary-oriented object detection in remote sensing images. By designing a multilayer-enhanced feature pyramid network and RH-head network, stable and accurate feature representations are obtained, and different orientations of objects are distinguished. Experimental results show that this method not only improves the detection accuracy but also enhances the visualized predicted results.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Boyong He, Bo Huang, Yue Shen, Liaoni Wu
Summary: In this paper, the challenging task of ship detection in aerial images is addressed through four proposed methods, including generating synthetic datasets, using balanced feature fusion structures, designing efficient anchor generation structures, and adopting IoU-based sampling methods. Experimental results demonstrate the significant improvement in ship object detection accuracy by combining these methods.
NEURAL COMPUTING & APPLICATIONS
(2022)
Article
Environmental Sciences
Zequn Sun, Chunning Meng, Jierong Cheng, Zhiqing Zhang, Shengjiang Chang
Summary: In this paper, the authors propose a multi-scale feature pyramid network (MS-FPN) for the detection and instance segmentation of marine ships in SAR images. The MS-FPN model uses a pyramid structure and consists of two modules, the atrous convolutional pyramid (ACP) module and the multi-scale attention mechanism (MSAM) module. The ACP module extracts both shallow and deep feature maps, crucial for describing multi-scale marine ships. The MSAM module adaptsively learns and selects important feature maps obtained from different scales, leading to improved accuracy in detection and segmentation.
Article
Engineering, Electrical & Electronic
Wenqing Zhao, Minfu Xu, Xingfu Cheng, Zhenbing Zhao
Summary: This article proposes an insulator recognition and fault detection model using techniques such as faster region convolutional neural network and feature pyramid networks to address the problem of insulator recognition in complex backgrounds. Experimental results show that the model achieves high accuracy and mean average precision in accurately recognizing and detecting faults in glass and composite insulators.
IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT
(2021)
Article
Engineering, Biomedical
Yanzhou Su, Jian Cheng, Chuqiao Zhong, Yijie Zhang, Jin Ye, Junjun He, Jun Liu
Summary: This paper proposes a method that optimizes both body and edge features simultaneously for the early detection and diagnosis of colorectal polyps. By introducing a Feature Decoupled Module (FDM), the method achieves superior performance on multiple datasets and demonstrates strong generalization ability.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Information Systems
Zhiqiang Zhang, Xin Qiu, Yongzhou Li
Summary: This paper proposes an improved method for FPN by introducing a multi-scale receptive fields extraction module, a feature constructor module, and an attention module to enhance the detection efficiency for objects of various scales and bridge the gap in content description and semantics between different layers. Experimental results show a significant improvement compared to FPN.
Article
Chemistry, Analytical
Quanzhi An, Zongxu Pan, Hongjian You
Article
Computer Science, Information Systems
Quanzhi An, Zongxu Pan, Hongjian You, Yuxin Hu
Summary: This paper mainly addresses several key issues in SAR image target detection, including the generality and specificity of features, detection under small sample conditions, as well as the feasibility and advantages of using an anchor-free rotatable detector.
