PRO-SSRGAN: stable super-resolution generative adversarial network based on parameter reconstructive optimization on Gaofen-5 remote-sensing images

Deep Learning image processing methods are gradually gaining popularity in a number of areas including medical imaging. Classification, segmentation, and denoising of images are some of the most demanded tasks. In this study, we aim at enhancing optic nerve head images obtained by Optical Coherence Tomography (OCT). However, instead of directly applying noise reduction techniques, we use multiple state-of-the-art image Super-Resolution (SR) methods. In SR, the low-resolution (LR) image is upsampled to match the size of the high-resolution (HR) image. With respect to image enhancement, the upsampled LR image can be considered as low quality, noisy image, and the HR image would be the desired enhanced version of it. We experimented with several image SR architectures, such as super-resolution Convolutional Neural Network (SRCNN), very deep Convolutional Network (VDSR), deeply recursive Convolutional Network (DRCN), and enhanced super-resolution Generative Adversarial Network (ESRGAN). Quantitatively, in terms of peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM), the SRCNN, VDSR, and DRCN significantly improved the test images. Although the ERSGAN showed the worst PSNR and SSIM, qualitatively, it was the best one.

Tropospheric Black Carbon (BC) aerosols are short lived positive radiative forcer with critical impacts on cardiovascular and pulmonary health. The MRI-ESM2(CMIP6) delivers the BC monthly global surface concentration from 1950-2100 at coarse special resolution of 1.875° × 1.875° which does not capture the city level BC hotspots at the global scale. The BC hotspots are essential for reducing the BC aerosols induced health burden, implementing air quality management policies and regional planning. The downscaling algorithm is an enhanced U-net with attention-based convolution neural network (Super Resolution Convolution Neural Network (SRCNN)). The SRCNN model executes downscaling of MRI-ESM2(CMIP6) BC monthly surface concentration with special resolution of 1.875° × 1.875° to NASA’s MERRA2 reanalysis BC monthly average surface concentration at a spatial resolution of 0.5° × 0.625°, thus achieving 3.6 times downscaling for identification city level BC-hotspots and cold spots at a global scale. The SRCNN model is trained on global monthly average BC surface concentration data from MRI-ESM2(CMIP6) and NASA’s MERRA2 reanalysis product. The model training is spread from 1980-2012(31 years) with validation from 2013-2016(4 years) and testing for 2017-2020(4 years). We have also examined the effects of Channel Based Attention Module (CBAM) with and without Residual Block (RB) and their effectiveness and efficacy in climate data downscaling with data-scarce condition. The training results showes that CBAM with RB based CNN outperforms then both (CNN without CBAM and without CBAM & RB) in the benchmarks such as stability, overfitting, validation losses etc. The training results for SRCNN (with CBAM and RB) shows a final validation losses of 0.0028, final R² value of 0.7162, final Pearson-r value of 0.8467 with Structural Similarity Index Measure (SSIM) at 0.9954. The SRCNN (with CBAM & RB) model testing reveals it performs exceptionally well in the identification of hotspots and cold-spots, with final testing RSME at 0.0015, final R2 at 0.88, final Pearson-r values at 0.94 and final SSIM at 0.99. Furthermore, testing outputs of SRCNN with attention module and residual blocks shows close fidelity with MERRA2 reanalysis vis-à-vis MRI-ESM2(CMIP6) at both seasonal and annual temporal resolution thus reducing systematic bias between ground truth and global climate models.

The development of deep learning and big data technology has introduced information and intelligent techniques to agricultural remote sensing estimation. The deep learning methods represented by Convolutional Neural Network (CNN) have abilities to extract the depth-dependent features of crop growth. In the field of crop yield estimation, the core challenge is to utilize CNN to extract the related information from remote sensing image. In this paper, we apply histogram dimensionality reduction and time series fusion to generate the input layer of CNN. In view of the data characteristics, the CNN network structure was designed to extract the features of winter wheat growth from multitemporal MODIS images for yield estimation in North China. The results showed that the estimated yield of winter wheat based on time-series remote sensing images is highly correlated with statistical data, with Pearson's r of 0.82, RMSE of 724.72 kg.hm−2. In the case of sufficient statistical data, the provincial model performs better. CNN is able to mine more relevant information and has higher robustness. It also provides a technical reference for estimating large-scale crop yield.

Normalized difference vegetation index (NDVI) carries valuable information related to the photosynthetic activity of vegetation and is essential for monitoring phenological changes and ecosystem dynamics. The medium to high spatial resolution satellite images from Landsat 8 and Sentinel 2 offer opportunities to generate dense NDVI time series at 10-m resolution to improve our understanding of the land surface processes. However, synergistic use of Landsat 8 and Sentinel 2 for generating frequent and consistent NDVI data remains challenging as they have different spatial resolutions and spectral response functions. In this letter, we developed an attentional super resolution convolutional neural network (ASRCNN) for producing 10-m NDVI time series through fusion of Landsat 8 and Sentinel 2 images. We evaluated its performance in two heterogeneous areas. Quantitative assessments indicated that the developed network outperforms five commonly used fusion methods [i.e., enhanced deep convolutional spatiotemporal fusion network (EDCSTFN), super resolution convolutional neural network (SRCNN), spatial and temporal adaptive reflectance fusion model (STARFM), enhanced STARFM (ESTARFM), and flexible spatiotemporal data fusion (FSDAF)]. The influence of the method selection on the fusion accuracy is much greater than that of the fusion strategy in blending Landsat-Sentinel NDVI. Our results illustrate the advantages and potentials of the deep learning approaches on satellite data fusion.

In this paper, we investigate a convolutional neural network (CNN) approach for light field (LF) super-resolution (SR). We are motivated by the assumption that image priors can be embedded into CNN, and both external and internal correlations are important in LFSR. The LF images are indeed natural images except for its angular resolution, so the external correlations help to super-resolve a single image from a collection of general images, whilst the internal correlations are essential to enhance a single view in LF with the details in the other views. Accordingly, we propose a two-stage CNN, where the two stages exploit the external and internal correlations, respectively. Moreover, to improve the generalization ability of the second-stage CNN for inter-view SR, we propose to align different views at patch level to compensate for the disparity that is essential to LFSR, thus the second stage is termed multi-patch fusion CNN. Experimental results demonstrate the superior performance of our two-stage CNN compared with the state-of-the-art CNN-based SR methods.

This study develops a deep reinforcement learning (DRL)-based ultrasound visual servoing (DBUVS) scheme for an autonomous robotic echocardiography system. The proposed approach aims to replace remote cardiac sonographers with a newly developed AI agent, Cardiac Sonographer Net (CS-Net), trained using DRL based on the Soft Actor-Critic (SAC) algorithm. To train CS-Net, a robotic echocardiography environment is simulated in GAZEBO simulator using a two-stage generative AI (GenAI) approach to produce highfidelity synthetic ultrasound images. In the first stage, multiple convolutional neural networks (CNNs) generate initial ultrasound images based on different parameter settings from a given probe pose. A fuzzy inference system (FIS) then fuses these images into a single low-quality representation. In the second stage, a super-resolution generative adversarial network with gradient penalty (SRGAN-GP) enhances image quality. Compared with low-quality images, the GenAI-based outputs show an 11.37 perceptual image patch similarity (LPIPS) and higher resolution (256W256 to 500W500 pixels), closely matching real ultrasound images. CS-Net is initially trained in the simulation environment and deployed on the real experimental robotic system with an ultrasound probe mounted on the end effector and a cardiac phantom for testing, using sim-to-real transfer learning. Experimental results demonstrate that the robotic echocardiography system powered by CS-Net performs autonomous scanning with higher accuracy and efficiency than the echocardiography remotely operated by the sonographer. Specifically, the system achieves faster convergence by reaching an image feature error norm of 0.176 in 25 seconds, compared with 0.253 in 50 seconds for remote operation.

Magnetic Resonance Imaging (MRI) provides a robust, non-invasive method for brain assessment and the detection of age-related biomarkers. However, due to hardware restrictions or the need for shorter acquisition times, clinical scans are often acquired at reduced spatial resolution, leading to degraded image quality that hampers tasks such as age prediction and other downstream analyses. To address this challenge, we propose a lightweight two-stage deep learning framework that combines super-resolution reconstruction with age regression. In the first stage, a Super-Resolution Convolutional Neural Network (SRCNN) enhances low-quality axial brain MRI slices, recovering fine structural details and improving overall image fidelity. The reconstructed images achieve a peak signal-to-noise ratio (PSNR) of 30.45 dB, a mean squared error (MSE) of 0.0010, and a structural similarity index (SSIM) of 0.9599 on the enhanced dataset. In the second stage, a specialized convolutional neural network leverages the super-resolved images to estimate biological age. Tested on the IXI dataset, the model demonstrates strong predictive accuracy, achieving a mean absolute error (MAE) of 2.35 years, a root mean squared error (RMSE) of 3.05 years, and a coefficient of determination (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex>) of 0.965. These results highlight the effectiveness of super-resolution as a preprocessing step for improving regression performance. The proposed lightweight and interpretable architecture illustrates the advantages of integrating high-fidelity reconstruction with predictive modeling in a unified pipeline, enabling improved brain MRI analysis and reliable age prediction.

To discuss the analysis and evaluation of highway landslides, the application of data mining methods combined with deep learning frameworks in geologic hazard evaluation and monitoring is explored preliminarily. On the premise of optimizing the processing of landslide images, first, the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) based on the natural statistical characteristics of the spatial domain is introduced, which is initially combined with Super‐Resolution Convolutional Neural Network (SRCNN). Then, the AlexNet is fine‐tuned and applied to highway landslide monitoring and surveying. Finally, an entropy weight gray clustering evaluation method based on data mining analysis is proposed, and the performances of several methods are verified. The results show that the average score of the BRISQUE algorithm in Image Quality Assessment (IQA) is above 0.9, and the average running time is 0.1523 s. The combination of BRISQUE and SRCNN can improve the image quality significantly. After fine‐tuning, the recognition accuracy of AlexNet for landslide images can reach about 80%. The evaluation method based on gray clustering can effectively determine the correlation between soil moisture content and slope angle and thereby be applied to the analysis and evaluation of highway landslides. The results are beneficial to the judgment and assessment of highway landslide conditions, which can be extended to research on other geologic hazards.

Despite advances in digital education making online oil painting lessons feasible, clarity and accuracy are still issues. Videos of low resolution or inadequate clarity make brushwork and texture, two essential elements of oil painting, difficult to discern. This work suggests an innovative deep learning-based online oil painting teaching framework (IDL-OOPTF) with picture super-resolution reconstruction to address this problem. The technique enhances real-time painting demos and student participation by utilizing a Convolutional Neural Network (CNN) model for semantic scene interpretation and an enhanced super-resolution generative adversarial network (SRGAN). Because of this invention, students can see finer details and color gradients in low-resolution television broadcasts. The proposed model outperforms baseline SR techniques and traditional interpolation on PSNR, SSIM, and customer satisfaction measures using a carefully chosen collection of oil painting pictures and instructional videos. Notable results include 35% more student engagement and 28% more detail recognition. Ultimately, this method enhances the pedagogical and aesthetic quality of online oil painting classes, addressing the issue of distant art education on a large scale. Customization using reinforcement learning and adaptive feedback are being developed.

Infrared images have been widely used in many research areas, such as target detection and scene monitoring. Therefore, the copyright protection of infrared images is very important. In order to accomplish the goal of image-copyright protection, a large number of image-steganography algorithms have been studied in the last two decades. Most of the existing image-steganography algorithms hide information based on the prediction error of pixels. Consequently, reducing the prediction error of pixels is very important for steganography algorithms. In this paper, we propose a novel framework SSCNNP: a Convolutional Neural-Network Predictor (CNNP) based on Smooth-Wavelet Transform (SWT) and Squeeze-Excitation (SE) attention for infrared image prediction, which combines Convolutional Neural Network (CNN) with SWT. Firstly, the Super-Resolution Convolutional Neural Network (SRCNN) and SWT are used for preprocessing half of the input infrared image. Then, CNNP is applied to predict the other half of the infrared image. To improve the prediction accuracy of CNNP, an attention mechanism is added to the proposed model. The experimental results demonstrate that the proposed algorithm reduces the prediction error of the pixels due to full utilization of the features around the pixel in both the spatial and the frequency domain. Moreover, the proposed model does not require either expensive equipment or a large amount of storage space during the training process. Experimental results show that the proposed algorithm had good performances in terms of imperceptibility and watermarking capacity compared with advanced steganography algorithms. The proposed algorithm improved the PSNR by 0.17 on average with the same watermark capacity.

ABSTRACTAnonymous password authenticated key exchange (APAKE) protocols allow the server to authenticate its clients without revealing their identities. In this paper, we first construct a basic protocol SAPAKE by using the homomorphic encryption scheme and an auxiliary memory device. Compared with the previous ones, SAPAKE is more suitable for those privacy‐sensitive applications (e.g., cloud computing) where reducing server payload and improving user experience are both essential. Furthermore, we refine SAPAKE by removing the use of the memory device to gain an enhanced extension SAPAKE+ without increasing the resources consumption. SAPAKE+ achieves better user‐friendliness than SAPAKE while it requires publishing more public parameters. Both of our protocols are practical due to their low (computation and communication) resources consumption and better user‐friendliness, and achieve provable security in the random oracle model. Copyright © 2012 John Wiley &amp; Sons, Ltd.

In this paper, we propose to address the downscaling of ocean remote sensing data using image super-resolution models based on deep learning, and more particularly Convolutional Neural Networks (CNNs). The goal of this study, for which we focus on satellite-derived Sea Surface Temperature (SST) data, is to evaluate the efficiency and the relevance of deep learning architectures applied to oceanographic remote sensing data. By using a CNN architecture, namely SRCNN (Super Resolution CNN), on a large-scale dataset of SST fields, we show that it allows a considerable gain in terms of PSNR compared to classical downscaling techniques. These results point out the relevance of deep learning models specifically trained for ocean remote sensing data and advocate for other applications to the reconstruction of high-resolution sea surface geophysical fields from multi-sensor satellite observations.

Low-resolution images can be reconstructed into high-resolution images using the Super-resolution Convolution Neural Network (SRCNN) algorithm. This study aims to improve the vehicle license plate number's recognition accuracy by generating a high-resolution vehicle image using the SRCNN. The recognition is carried out by two types of character recognition methods: Tesseract OCR and SPNet. The training data for SRCNN uses the DIV2K dataset consisting of 900 images, while the training data for character recognition uses the Chars74 dataset. The high-resolution images constructed using SRCNN can increase the average accuracy of vehicle license plate number recognition by 16.9 % using Tesseract and 13.8 % with SPNet.

This article presents a frame of research for information retrieval from remote sensing (RS) images, based on a new concept developed at the German Aerospace Center, DLR Oberpfaffenhofen in collaboration with the Swiss Federal Institute of Technology, ETH Zurich. The newly developed technology emphasises the user access to the intrinsic information in RS images and on information instead of data dissemination. Thus the users are enabled to enhance the robustness of their decisions and to extend the range of applications.

The growth of rubber trees always suffer from blight disease that can be detected on the leaves. Several studies show that an early detection of blight disease may contribute to a positive turnout recovery rate. In this paper, a mobile app-based microscopic leaf imaging disease classification is proposed to detect infected leaves that provides a probability of disease types, geo-tagging location, and smart reporting with recovery stage to assist productivity workflow. Super Resolution Generative Adversarial Network is applied to upscale a low resolution microscopic leaf imaging while introducing finer texture detail. After super resolution reconstruction, a convolution neural network classifier is performed to classify disease groups with an improved accuracy. The diagnostic solution is designed on the Azure cloud computing service to manage the plant disease database, perform reinforcement learning, host web application, secure authentication and display valuable insight of recovery.