Efficient Residual Dense Block Search for Image Super-Resolution

Although remarkable progress has been made on single image super-resolution due to the revival of deep convolutional neural networks, deep learning methods are confronted with the challenges of computation and memory consumption in practice, especially for mobile devices. Focusing on this issue, we propose an efficient residual dense block search algorithm with multiple objectives to hunt for fast, lightweight and accurate networks for image super-resolution. Firstly, to accelerate super-resolution network, we exploit the variation of feature scale adequately with the proposed efficient residual dense blocks. In the proposed evolutionary algorithm, the locations of pooling and upsampling operator are searched automatically. Secondly, network architecture is evolved with the guidance of block credits to acquire accurate super-resolution network. The block credit reflects the effect of current block and is earned during model evaluation process. It guides the evolution by weighing the sampling probability of mutation to favor admirable blocks. Extensive experimental results demonstrate the effectiveness of the proposed searching method and the found efficient super-resolution models achieve better performance than the state-of-the-art methods with limited number of parameters and FLOPs.

Download Full-text

Deep Residual Squeeze and Excitation Network for Remote Sensing Image Super-Resolution

Remote Sensing ◽

10.3390/rs11151817 ◽

2019 ◽

Vol 11 (15) ◽

pp. 1817 ◽

Cited By ~ 8

Author(s):

Jun Gu ◽

Xian Sun ◽

Yue Zhang ◽

Kun Fu ◽

Lei Wang

Keyword(s):

Remote Sensing ◽

State Of The Art ◽

Super Resolution ◽

Remote Sensing Image ◽

Single Image ◽

Remote Sensing Images ◽

Deep Convolutional Neural Networks ◽

Current Block ◽

Image Super Resolution ◽

Single Image Super Resolution

Recently, deep convolutional neural networks (DCNN) have obtained promising results in single image super-resolution (SISR) of remote sensing images. Due to the high complexity of remote sensing image distribution, most of the existing methods are not good enough for remote sensing image super-resolution. Enhancing the representation ability of the network is one of the critical factors to improve remote sensing image super-resolution performance. To address this problem, we propose a new SISR algorithm called a Deep Residual Squeeze and Excitation Network (DRSEN). Specifically, we propose a residual squeeze and excitation block (RSEB) as a building block in DRSEN. The RSEB fuses the input and its internal features of current block, and models the interdependencies and relationships between channels to enhance the representation power. At the same time, we improve the up-sampling module and the global residual pathway in the network to reduce the parameters of the network. Experiments on two public remote sensing datasets (UC Merced and NWPU-RESISC45) show that our DRSEN achieves better accuracy and visual improvements against most state-of-the-art methods. The DRSEN is beneficial for the progress in the remote sensing images super-resolution field.

Download Full-text

DEEP RESIDUAL LEARNING FOR SINGLE-IMAGE SUPER-RESOLUTION OF MULTI-SPECTRAL SATELLITE IMAGERY

ISPRS Annals of Photogrammetry Remote Sensing and Spatial Information Sciences ◽

10.5194/isprs-annals-iv-2-w7-189-2019 ◽

2019 ◽

Vol IV-2/W7 ◽

pp. 189-196

Author(s):

L. Wagner ◽

L. Liebel ◽

M. Körner

Keyword(s):

Remote Sensing ◽

Satellite Imagery ◽

Network Architecture ◽

Super Resolution ◽

Superior Performance ◽

Optical Remote Sensing ◽

Single Image ◽

Image Super Resolution ◽

Single Image Super Resolution ◽

The Individual

Abstract. Analyzing optical remote sensing imagery depends heavily on their spatial resolution. At the same time, this data is adversely affected by fixed sensor parameters and environmental influences. Methods for increasing the quality of such data and concomitantly optimizing its information content are, thus, in high demand. In particular, single-image super-resolution (SISR) approaches aim to achieve this goal solely by observing the individual images.We propose to adapt a generic deep residual neural network architecture for SISR to deal with the special properties of remote sensing satellite imagery, especially taking into account the different spatial resolutions of individual Sentinel-2 bands, i.e., ground sampling distances of 20&thinsp;m and 10&thinsp;m. As a result, this method is able to increase the perceived resolution of the 20&thinsp;m channels and mesh all spectral bands. Experimental evaluation and ablation studies on large datasets have shown superior performance compared to the state-of-the-art and that the model is not bound by its capacity.

Download Full-text

RBDN: Residual Bottleneck Dense Network for Image Super-Resolution

10.21203/rs.3.rs-309485/v1 ◽

2021 ◽

Author(s):

Zeyu An ◽

Junyuan Zhang ◽

Ziyu Sheng ◽

Xuanhe Er ◽

Junjie Lv

Keyword(s):

Network Architecture ◽

Super Resolution ◽

Dense Network ◽

Single Image ◽

Generative Adversarial Network ◽

Adversarial Network ◽

Network Building ◽

Image Super Resolution ◽

Feature Values ◽

Single Image Super Resolution

Abstract Recent studies have shown that Super-Resolution Generative Adversarial Network (SRGAN) can significantly improve the quality of single-image super-resolution. However, the existing SRGAN approaches also have drawbacks, such as inadequate of features utilization, huge number of parameters and poor scalability. To further enhance the visual quality, we thoroughly study three key components of SRGAN: network architecture, adversarial loss and perceptual loss, and propose a DenseNet with Residual-in-Residual Bottleneck Block (RRBB) named as Residual Bottleneck Dense Network (RBDN) for single-image super-resolution. In particular, RBDN combines ResNet and DenseNet with different roles, in which ResNet refines feature values by addition and DenseNet memorizes feature values by concatenation. Specifically, the DenseNet adopts the Bottleneck structure to reduce the network parameters and improve the convergence rate. In addition, the proposed RRBB, as the basic network building unit, removes the batch normalization (BN) layer and employs the ELU function to reduce the opposite effects in the absence of BN. In this way, RBDN can enjoy the merits of the sufficient feature value refined by residual groups and the refined feature value memorized by dense connections, thus achieving better performance compared with most current residual blocks.

Download Full-text

HNSR: Highway Networks Based Deep Convolutional Neural Networks Model for Single Image Super-Resolution

2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) ◽

10.1109/icassp.2018.8462007 ◽

2018 ◽

Cited By ~ 2

Author(s):

Ke Li ◽

Bahetiyaer Bare ◽

Bo Yan ◽

Bailan Feng ◽

Chunfeng Yao

Keyword(s):

Neural Networks ◽

Convolutional Neural Networks ◽

Super Resolution ◽

Single Image ◽

Deep Convolutional Neural Networks ◽

Highway Networks ◽

Image Super Resolution ◽

Single Image Super Resolution

Download Full-text

Triple-Attention Mixed-Link Network for Single-Image Super-Resolution

Applied Sciences ◽

10.3390/app9152992 ◽

2019 ◽

Vol 9 (15) ◽

pp. 2992 ◽

Cited By ~ 1

Author(s):

Xi Cheng ◽

Xiang Li ◽

Jian Yang

Keyword(s):

Receptive Fields ◽

Super Resolution ◽

Feature Representation ◽

Single Image ◽

Deep Convolutional Neural Networks ◽

Flow Topology ◽

Image Super Resolution ◽

Representational Power ◽

Pass Through ◽

Single Image Super Resolution

Single-image super-resolution is of great importance as a low-level computer-vision task. Recent approaches with deep convolutional neural networks have achieved impressive performance. However, existing architectures have limitations due to the less sophisticated structure along with less strong representational power. In this work, to significantly enhance the feature representation, we proposed triple-attention mixed-link network (TAN), which consists of (1) three different aspects (i.e., kernel, spatial, and channel) of attention mechanisms and (2) fusion of both powerful residual and dense connections (i.e., mixed link). Specifically, the network with multi-kernel learns multi-hierarchical representations under different receptive fields. The features are recalibrated by the effective kernel and channel attention, which filters the information and enables the network to learn more powerful representations. The features finally pass through the spatial attention in the reconstruction network, which generates a fusion of local and global information, lets the network restore more details, and improves the reconstruction quality. The proposed network structure decreases 50% of the parameter growth rate compared with previous approaches. The three attention mechanisms provide 0.49 dB, 0.58 dB, and 0.32 dB performance gain when evaluating on Set5, Set14, and BSD100. Thanks to the diverse feature recalibrations and the advanced information flow topology, our proposed model is strong enough to perform against the state-of-the-art methods on the benchmark evaluations.

Download Full-text

Downscaling of Precipitation Forecasts Based on Single Image Super-Resolution

10.5194/egusphere-egu2020-8533 ◽

2020 ◽

Cited By ~ 1

Author(s):

Yan Ji ◽

Xiefei Zhi ◽

Ye Tian ◽

Ting Peng ◽

Ziqiang Huo ◽

...

Keyword(s):

Spatial Information ◽

Daily Precipitation ◽

Regional Scale ◽

Super Resolution ◽

Ensemble Prediction ◽

Generative Adversarial Networks ◽

Single Image ◽

Deep Convolutional Neural Networks ◽

Image Super Resolution ◽

Single Image Super Resolution

High spatial resolution weather forecasts that capture regional-scale dynamics are important for natural hazards prevention, especially for the regions featured with large topographical variety and local climate. While deep convolutional neural networks have made great progress in single image super-resolution (SR) which learns mapping relationship between low- and high- resolution images, limited efforts have been made to explore the potential of downscaling in this way. In the study, three advanced SR deep learning frameworks including Super-Resolution Convolutional Neural Network (SRCNN), Super-Resolution Generative Adversarial Networks (SRGAN) and Enhanced Deep residual networks for Super-Resolution (EDSR) are proposed for downscaling forecasts of daily precipitation in southeast China (100&#176;E -130&#176;E, 15&#176;N -35&#176;N). The SR frameworks are designed to improve the horizontal resolution of daily precipitation forecasts from raw 1/2 degrees (~50km) to 1/4 degrees (~25km) and 1/8 degrees (~12.5km), respectively. For comparison, Bias Correction Spatial Disaggregation (BCSD) as a traditional SD method is also performed under the same framework. The precipitation forecasts used in our work are obtained from different Ensemble Prediction Systems (EPSs) including ECMWF, NCEP and JMA which are provided by TIGGE datasets. A group of metrics have been applied to assess the performance of the three SR models, including Root Mean Square Error (RMSE), Anomaly Correlation Coefficient (ACC) and Equitable Threat Score (ETS). Results show that three SR models can effectively capture the detailed spatial information of local precipitation that is ignored in global NWPs. Among the three SR models, EDSR obtains the optimum results with lower RMSE and higher ACC which shows better downscaling skills. Furthermore, the SR downscaling methods can be extended to the statistical downscaling for other predictors as well.

Download Full-text

Deep fractal residual network for fast and accurate single image super resolution

Neurocomputing ◽

10.1016/j.neucom.2019.09.093 ◽

2020 ◽

Vol 398 ◽

pp. 389-398 ◽

Cited By ~ 1

Author(s):

Yanghao Zhou ◽

Jianfeng Dong ◽

Yubin Yang

Keyword(s):

Super Resolution ◽

Single Image ◽

Residual Network ◽

Image Super Resolution ◽

Single Image Super Resolution

Download Full-text

Real-Time Environment Monitoring Using a Lightweight Image Super-Resolution Network

International Journal of Environmental Research and Public Health ◽

10.3390/ijerph18115890 ◽

2021 ◽

Vol 18 (11) ◽

pp. 5890

Author(s):

Qiang Yu ◽

Feiqiang Liu ◽

Long Xiao ◽

Zitao Liu ◽

Xiaomin Yang

Keyword(s):

Deep Learning ◽

Real Time ◽

Super Resolution ◽

Model Complexity ◽

Practical Application ◽

Single Image ◽

Feature Maps ◽

Benchmark Datasets ◽

Image Super Resolution ◽

Single Image Super Resolution

Deep-learning (DL)-based methods are of growing importance in the field of single image super-resolution (SISR). The practical application of these DL-based models is a remaining problem due to the requirement of heavy computation and huge storage resources. The powerful feature maps of hidden layers in convolutional neural networks (CNN) help the model learn useful information. However, there exists redundancy among feature maps, which can be further exploited. To address these issues, this paper proposes a lightweight efficient feature generating network (EFGN) for SISR by constructing the efficient feature generating block (EFGB). Specifically, the EFGB can conduct plain operations on the original features to produce more feature maps with parameters slightly increasing. With the help of these extra feature maps, the network can extract more useful information from low resolution (LR) images to reconstruct the desired high resolution (HR) images. Experiments conducted on the benchmark datasets demonstrate that the proposed EFGN can outperform other deep-learning based methods in most cases and possess relatively lower model complexity. Additionally, the running time measurement indicates the feasibility of real-time monitoring.

Download Full-text