Laplacian Feature Pyramid Network for Object Detection in VHR Optical Remote Sensing Images

Zhang, Wenhua; Jiao, Licheng; Li, Yuxuan; Huang, Zhongjian; Wang, Haoran

doi:10.1109/tgrs.2021.3072488

Cited by 32 publications

(16 citation statements)

References 51 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After generating the image pyramid from the original image, [23] put components into two modules for highquality transfer, a drafting module that transforms artistic style from low-resolution image and a revision module that refines the transformed image with a high-resolution image. On top of these, there are applications for diverse problems such as object detection [42], image compression [37]. These researches show a possibility of a connection between LP and deep learning, presenting satisfactory results.…”

Section: Strengths Of Lp and Applications On Neural Netmentioning

confidence: 99%

Laplacian Pyramid-like Autoencoder

Han

Hur²,

Hur³

2022

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

In this paper, we develop the Laplacian pyramid-like autoencoder (LPAE) by adding the Laplacian pyramid (LP) concept widely used to analyze images in Signal Processing. LPAE decomposes an image into the approximation image and the detail image in the encoder part and then tries to reconstruct the original image in the decoder part using the two components. We use LPAE for experiments on classifications and super-resolution areas. Using the detail image and the smaller-sized approximation image as inputs of a classification network, our LPAE makes the model lighter. Moreover, we show that the performance of the connected classification networks has remained substantially high. In a superresolution area, we show that the decoder part gets a high-quality reconstruction image by setting to resemble the structure of LP. Consequently, LPAE improves the original results by combining the decoder part of the autoencoder and the super-resolution network.

show abstract

Section: Strengths Of Lp and Applications On Neural Netmentioning

confidence: 99%

Laplacian Pyramid-like Autoencoder

Han

Hur²,

Hur³

2022

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

show abstract

“…RSI target detection is the accurate identification of classes and locations of ground objects in high altitude. Accurately locating ships and vehicles in images captured under high-altitude [7][8][9] conditions is one of these tasks. Recently, RSI detection has become increasingly efficient and accurate due to the continuous application of advanced methods such as large kernel computation and attention mechanisms.…”

Section: Introductionmentioning

confidence: 99%

A remote sensing target detection model based on lightweight feature enhancement and feature refinement extraction

Guo,

Zhou,

Guo

et al. 2024

Preprint

View full text Add to dashboard Cite

Remote sensing image (RSI) target detection methods based on traditional multi scale feature fusion (MSFF) have achieved great success. However, the traditional MSFF method significantly increases the computational cost during model training and inference, and the simple fusion operation may lead to the semantic confusion of the feature map, which cannot realize the refined extraction of features by the model. In order to reduce the computational effort associated with the MSFF operation and to enable the features in the feature map to present an accurate, fine-grained distribution, we propose a single-stage detection model(RS-YOLO). Our main additions to RS-YOLO are a computationally smaller and faster QS-E-ELEN (Quick and Small E-ELEN) module and a feature refinement extraction (FRE) module. In the QS-E-ELEN module We utilize QSBlock,jump-join, and convolution operations to fuse features on different scales and reduce the computational effort of the model by exploiting the similarity of the RSI feature map channels. In order for the model to better utilize the enhanced features, FRE makes the feature mapping of the target to be detected in the RSI accurate and refined. By conducting experiments on the popular NWPU-VHR- 10 and SSDD datasets, we derive results that show that RS-YOLO outperforms most mainstream models in terms of the trade-off between accuracy and speed. Specifically, in terms of accuracy, it improves 1.6% and 1.7% compared to the current state-of-the-art models, respectively. At the same time, RS-YOLO reduces the number of parameters and computational effort.

show abstract

“…Given the dependency of directional WFS formulation on both spreading function and its boundary conditions, which includes the effects of sea states [84], [108], [112], no ROM model generated thus far outperforms others in terms of representing reference roughness pattern and texture characteristics [45], [49], [56], [63], [65], [77], [88], [89], [93], [99], [103], [108], [113], [117], [120]. To address this limitation, a multi-scale transform domain (MTD) fusion method is proposed [124]- [127], incorporating a convolutional neural network (CNN) [128], [129], enabling the reconstruction of a fused roughness model independent of spreading functions as a reference ROM [50], [80], [103], [129], and facilitating the generation of an optimized SAR raw data [59], [60], [73], [74], [77], [80], [83]- [87], [93], [106], [107], [119], [121].…”

Section: Introductionmentioning

confidence: 99%

Directional Wave Scattering Distribution Modes Analysis and Synthesis of Random Ocean Media Roughness for SAR Electromagnetic Interactions Using Feature Fusion in Dynamic Sea States: A Survey

Shahrezaei,

Kim

2024

IEEE Access

View full text Add to dashboard Cite

Ocean waves have long been a research topic, and numerous formulas of the ocean wave spectrum have been widely developed to provide new prospects for ocean experiments and the advancement of radar probing. Nonetheless, the wave spectra developed by researchers fall short of the standards set by remote sensing specialists, mainly due to the limitation of capturing the surface roughness influenced by both short and long waves, prompting ongoing efforts to develop a model covering a diverse scale of wavenumbers in the absence of a generally recognized reference formula. In response, a standard two-scale formulation of wave frequency spectra (WFS) was introduced, where wave height and spectral peak period are determined by sea state. The proposed composite WFS holds the potential to incorporate directional spreading, contributing to the angular distribution of ocean wave energy in the form of directional WFS, making it applicable for ocean modeling. In an effort to investigate directionality effects, an array of well-established spreading functions, including cosine-squared, half-cosine 2s, parameterized half-cosine 2s, hyperbolic secant-squared, and composite structured functions, has been developed here for numerical modeling of random ocean media (ROM) surface roughness and synthesized for their spectral scattering distribution mode, encompassing scattering pattern, scattering orientation, and fractal roughness properties. Nonetheless, the generated ROM models, each varying due to inherent limitations in directional WFS formulation and numerical approximations, demonstrate indeterminacy and unpredictability in surface features, posing challenges for accurately synthesizing ROM roughness patterns. These challenges intensify under varying sea states and different directional WFS formulas, leading to a situation where no single synthesized composite ROM model consistently outperforms the others, rendering them imprecise frameworks for analyzing roughness patterns and investigating texture electromagnetic interactions within the realm of remote sensing. As an approach, a pattern-sensitive fusion method is proposed, employing a multi-scale transform domain (MTD) fusion scheme that leverages the learning potential of a deep super resolution network. The objective is to fuse the reconstructed ROM roughness models, generating an optimal roughness while maintaining their scattering pattern, scattering orientation, and dominant directionality, pivotal for texture consistency and, consequently, the backscattering properties from the synthetic aperture radar (SAR) viewpoint. To validate the reliability of ROM modeling and its roughness synthesis, including the texture fusion and raw data generation, a comprehensive objective quality assessment technique is utilized. These assessments demonstrate the complete consistency of the simulation results with the underlying spectral theory, highlighting their potential contribution to projects related to ocean radar probing and remote sensing.INDEX TERMS Directional wave frequency spe...

show abstract

Laplacian Feature Pyramid Network for Object Detection in VHR Optical Remote Sensing Images

Cited by 32 publications

References 51 publications

Laplacian Pyramid-like Autoencoder

Laplacian Pyramid-like Autoencoder

A remote sensing target detection model based on lightweight feature enhancement and feature refinement extraction

Directional Wave Scattering Distribution Modes Analysis and Synthesis of Random Ocean Media Roughness for SAR Electromagnetic Interactions Using Feature Fusion in Dynamic Sea States: A Survey

Contact Info

Product

Resources

About