Image noise, often resulting from disturbances during image acquisition or imperfections in the imaging device, notably degrades the quality of digital images. The challenge of removing this noise has been addressed through various techniques, from traditional filtering and prior-based methods to more recent deep learning approaches. In this paper, we introduce DRSFANet, an advanced dual-path convolutional neural network (CNN) specifically designed to tackle both synthetic Additive White Gaussian Noise (AWGN) and real-world noise. DRSFANet incorporates several innovative components: a residual feature extraction module (FEB) equipped with dilated convolutional layers to enhance the receptive field and mitigate gradient vanishing issues, and novel attention modules—Frequency-Plane Attention Block (FPAB) and Residual Attention Block (RAB)—which improve feature extraction in both frequency and spatial domains. Furthermore, the model features a downsampling (DS) block that effectively consolidates essential features prior to their integration into subsequent network stages. Comprehensive experimental evaluations reveal that DRSFANet outperforms several state-of-the-art denoising methods, demonstrating superior performance in both synthetic and real datasets through rigorous quantitative and qualitative analysis.

The interpretation of remote sensing images remains a significant challenge due to their complex, information-rich nature. Current Remote Sensing Visual Question Answering (RSVQA) techniques have been a step forward towards building intelligent analysis systems for remote sensing images. However most existing RSVQA models rely on already existing deep learning models for their representation (visual and language feature extraction) and fusion (combining the extracted features) modules, which poses a limitation to their performance. To address these limitations, this paper introduces a novel Remote Sensing Visual Question Answering (RSVQA) approach that leverages state-of-the-art components with an innovative architecture to advance interactive remote sensing analysis. The proposed model features a novel dual-layer visual attention mechanism in the Representation module to process intricate features and capture regional relationships alongside processing the overall features. The Fusion module employs a unique attention-based design, combining both self-attention and mutual attention, to integrate these features into a unified vector representation. Finally, the Answering module utilizes a refined Multi-Layer Perceptron classifier for precise response generation. Evaluations on RSVQA benchmarks demonstrate the system’s superiority over existing methods, marking a significant step forward in remote sensing analytics.

Persistent phenomena of transmitted frequency interference (after reflecting off the target location) lead to the introduction of random speckle distributions in the raw data collected by synthetic aperture radar (SAR) sensors. The quality of the acquired images is thus degraded significantly due to the undesired speckle which creates a granular cover across the visual. Numerous techniques have been proposed in the literature which aim to remove this undesired speckle component. However, the objective of removing speckle while preserving minute structural and textural information captured by the raw data still remains an open problem. This letter proposes a unique SAR despeckling approach that uses parallel convolutional encoder (PCE) technique which captures highly effective feature components at various processing levels. In addition, the residual-based encoder module is structured in a way so that it can capture the interdependence among the parallelly extracted feature components. Optimal utilization of the proposed network structure enables efficient analysis and subsequent removal of the speckle components while retaining minute details captured by the raw data. Experimental results across both simulated and real SAR data strongly support the proposed model’s superiority over various classical and state-of-the-art approaches described in the literature.

With the increasing demand of capturing and processing the visual data of Earth’s surface, synthetic aperture radar (SAR) technology has been widely accepted as the most preferred solution by various organizations. But a major limitation in processing SAR data is its inherent contamination with unwanted random granular interference known as “speckle.” The removal of these unwanted speckle components in order to extract a clear SAR visual, a process known as “despeckling,” forms an important preprocessing task. In this article, we propose a novel model-based technology for the despeckling of SAR visuals contaminated with speckle. The proposed model uses a generative adversarial network (GAN) module for extracting the unwanted speckle component from the input SAR data, which is then analyzed by a convolutional neural network (CNN) module for predicting the noise level (look). The predicted noise level serves as a thresholding parameter for the gradient information extracted from the input SAR data. This eliminates the gradient captured due to the speckle granular effect in the input data. Later the input SAR data, along with the extracted noise and the processed gradient, is fed to a deep dilated CNN-based restoration module to generate a clean SAR visual. Rather than traditional learning of either the residual noisy component or the clean data, the proposed despeckling model learns the pattern in which these noisy components degrade the original data and its gradient information. This methodology, in turn, significantly improves the despeckling performance, when compared with other existing technology in the literature.

Hyperspectral imaging (HSI) is the procedure of acquiring a scene over a wide range of electromagnetic spectrum for the purpose of detailed analysis and prediction. The occurrence of noise during the acquisition procedure, however, poses a limitation on this imaging system. Noise in HSI is classified as a mixture of Gaussian and impulse noise statistics, and noise removal or denoising forms an integral part of this imaging system. In this paper, we consider the problem of removing this mixed Gaussian-impulse noise from HSI data-sets by formulating a joint optimization problem based on the maximuma posteriori(MAP) estimates for Gaussian and impulse noise distributions. The proposed method is then solved using an efficient minimization strategy realizied through half-quadratic split. Extensive experimentation on synthetic and real HSI data-sets corroborate the effectiveness of the proposed denoising technique.

Acquired speckle noise degrades the quality of synthetic aperture radar (SAR) images which affect their further processing and analysis. In this paper we propose a sharp image feature based SAR denoising network called SIFSDNet. This network uses a image sharpening block, which amplifies the feature content of the input noisy image. Features of this sharpened image is then extracted and concatenated along with the features of the input noisy image to get a combined feature map. The despeckled SAR image is then reconstructed from the combined feature map. The proposed method supersedes state-of-the-art classical as well as deep learning based SAR denoising methods, both in terms of visual results as well as quantitative analysis. We have shared the code of SIFSDNet at https://github.com/RTSIR/SIFSDNet.

Many deep learning networks focus on the task of Gaussian denoising by processing images on a fixed scale or multiple scales using convolution and deconvolution. In certain cases, excessive scaling applied in the network results in the loss of image details. Sometimes, the usage of deeper convolutional networks results in the loss of network gradient. In this paper, to overcome both the problems, we propose a multi-scale pixel attention-based blind Gaussian denoiser network that utilizes a combination of important features at five different scales. The proposed network performs blind Gaussian denoising in the sense that it does not need any prior information about noise. It comprises a central multi-scale pixel attention block together with dilated convolutional layers and skip connections that help in utilizing the full receptive field of the first convolutional layer to the last convolutional layer and is based on residual architecture for propagating high-level information easily in the network. We have provided the code of the proposed technique at https://github.com/RTSIR/MSPABDN.

Synthetic aperture radar (SAR) images are mainly corrupted by speckle noise, which needs to be removed for further processing. In this letter, we propose an attention and gradient-based SAR denoising network (AGSDNet) to remove speckle noise from SAR images while preserving finer details. In the proposed network, gradient information of the noisy image is first concatenated with its features in order to increase the feature information content (map). An intermediate feature denoising block (FDB) is then employed to reduce noise from this feature map. Finally, two attention blocks, designed and deployed, in the network focus on preserving the more informative features in the image thereby generating a feature preserved denoised image. The proposed network is compared with several classical and deep learning-based SAR denoising methods to demonstrate its superiority in terms of qualitative as well as quantitative measures. We have provided the training and testing code of AGSDNet athttps://github.com/RTSIR/AGSDNet.

The authors present a proximal approach to hyperspectral image denoising adapted to the mixed noise behaviour of hyperspectral data; named hyperspectral image proximal denoiser ( HSIProxDenoiser ). A combination of Gaussian‐impulse noise has been handled under maximum a posteriori framework using two data fidelity terms. They have incorporated prior information about the data in the form of two regularisation terms, namely Tikhonov–Miller (TM) and total variation (TV). Since TV possesses feature selection capability by setting some of the coefficients to zero, it works well when there are a small number of significant features. On the other hand, TM works well if there are a large number of similar features. Hence, including both regularisation terms can help achieve the desired denoising performance. The resultant optimisation problem is solved using a variant of primal‐dual hybrid gradient by splitting the former into different functions and calculating their proximal operators individually. Experimental results over both synthetic as well as real hyperspectral image data validate the potential of the proposed technique both visually and in terms of quantitative metrics.

A new denoising technique is proposed in this study for synthetic aperture radar (SAR) images corrupted by speckle noise. The authors method extract informative features from a noisy speckled image, and then a denoised version of this image is estimated from the informative gradients, which are restricted to the features of this image. The technique of extracting features is designed on the framework of multifractal formalism followed by a reconstruction technique for the informative gradients based on the total variational (TV) regularisation framework. Experimental results demonstrate that the proposed approach is able to retain the finer details of the original image while removing noise. The superiority of the proposed approach is manifested qualitatively and quantitatively on comparing with state‐of‐the‐art denoising techniques.

In this letter, we propose a speckle removal denoising algorithm for synthetic aperture radar (SAR) images. The approach is based on the concept of extracting informative feature (based on the concept of multifractal decomposition of signals) from a speckle-induced SAR image and then estimating a noise-free image from the gradients restricted to those features. The experimental results show that the proposed technique not only improves the visual quality of the SAR images but also effectively preserves their texture. Comparison with the classical and state-of-the-art denoising techniques shows the advantages of the proposed scheme, both visually and quantitatively.

Turbulence in the Earth's atmosphere interferes with the propagation of planar wavefronts from outer space, resulting in a phase-distorted nonplanar wavefront. This phase distortion is responsible for the refractive blurring of images accounting to the loss in spatial resolution power of ground-based telescopes. The technology widely used to remove this phase distortion is adaptive optics (AO). In AO, an estimate of the distorted phase is provided by a wavefront sensor (WFS) in the form of low-resolution slope measurements of the wavefront. The estimate is then used to create a corrected wavefront that (approximately) removes the phase distortion from the incoming wavefronts. Phase reconstruction from WFS measurements is done by solving large linear systems, followed by interpolating the low-resolution phase to its desired high resolution. In this paper, we propose an alternate technique to wavefront phase reconstruction using concepts derived from the microcanonical multiscale formalism, which is a specific approach to multifractality. We take into account an a priori information of the wavefront phase, provided by the multifractal exponents. Then, through the framework of multiresolution analysis and wavelet transform, we address the problem of phase reconstruction from low-resolution WFS measurements. Comparison, in terms of reconstruction quality, with classical techniques in AO proves the superiority of our approach.