debajyoti dhar

Followers

Following

Public Views

Interests

Uploads

Papers by debajyoti dhar

SPRINT: Spectra Preserving Radiance Image Fusion Technique using holistic deep edge spatial attention and Minnaert guided Bayesian probabilistic model

Signal Processing: Image Communication

DeepSWIR: A Deep Learning Based Approach for the Synthesis of Short-Wave InfraRed Band using Multi-Sensor Concurrent Datasets

arXiv (Cornell University), May 7, 2019

Convolutional Neural Network (CNN) is achieving remarkable progress in various computer vision ta... more Convolutional Neural Network (CNN) is achieving remarkable progress in various computer vision tasks. In the past few years, the remote sensing community has observed Deep Neural Network (DNN) finally taking off in several challenging fields. In this study, we propose a DNN to generate a predefined High Resolution (HR) synthetic spectral band using an ensemble of concurrent Low Resolution (LR) bands and existing HR bands. Of particular interest, the proposed network, namely DeepSWIR, synthesizes ShortWave InfraRed (SWIR) band at 5m Ground Sampling Distance (GSD) using Green (G), Red (R) and Near InfraRed (NIR) bands at both 24m and 5m GSD, and SWIR band at 24m GSD. To our knowledge, the highest spatial resolution of commercially deliverable SWIR band is at 7.5m GSD. Also, we propose a Gaussian feathering based image stitching approach in light of processing large satellite imagery. To experimentally validate the synthesized HR SWIR band, we critically analyse the qualitative and quantitative results produced by DeepSWIR using state-of-the-art evaluation metrics. Further, we convert the synthesized DN values to Top Of Atmosphere (TOA) reflectance and compare with the corresponding band of Sentinel-2B. Finally, we show one real world application of the synthesized band by using it to map wetland resources over our region of interest.

Download

A novel country-level integrated image mosaic system using optical remote sensing imagery

Earth Science Informatics

Feature based remote sensing image registration techniques: a comprehensive and comparative review

International Journal of Remote Sensing

NL-FFC: Non-Local Fast Fourier Convolution for Image Super Resolution

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

Self-Supervised Variable Rate Image Compression using Visual Attention

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

An Automated Feature Tracking Based Algorithm for Estimation of Ice Flow and Glacier Surface Velocity Using Resourcesat-2A Data for Himalayan Terrain

2019 Fifth International Conference on Image Information Processing (ICIIP)

Determination of ice flow and glacier surface velocity plays a pivotal role in the study of glaci... more Determination of ice flow and glacier surface velocity plays a pivotal role in the study of glacial geography, its sensitivity to climate change and its impact on water cycle of the region. This study becomes of utmost importance in case of Himalayan Cryosphere, which is a source of water to many areas in Asia. In this paper, we propose an algorithm for optical remote sensing multispectral data which can automatically calculate the ice flow velocity and further help other glacier related studies. An ortho-rectified image pair of Resourcesat - 2A LISS-3 (Linear Imaging and Self Scanning) sensor is taken as an input, which is further subjected to a cloud mask and snow feature enhancement technique to improve the visibility of the requisite area. This in turn is used as an input to the first level Principal component analysis (PCA). Further a common raster is extracted and then moving chip based sub-pixel correlation is performed to find the ice shift. Post processing steps are used to remove spurious matches and false positives. In the analysis we observe variations in velocities during the winter and summer season, further fluctuations were also observed in velocities for different zones of glaciers. Average glacier surface velocities ranging from 30my−1 to 110my−1 are observed. This study shows that optical image based correlation is suitable and efficient for measurement of ice flow in the fast moving glaciers of the Himalayas.

An Efficient Approach for Geo-location Accuracy Improvement of Legacy IRS-1C/1D Data Products

2019 IEEE Recent Advances in Geoscience and Remote Sensing : Technologies, Standards and Applications (TENGARSS), 2019

IRS-1C/1D remote sensing datasets from Indian Space Research organization (ISRO) is of significan... more IRS-1C/1D remote sensing datasets from Indian Space Research organization (ISRO) is of significant importance till now for change detection analysis in the field of agriculture growth, wetland mapping, urban sprawl monitoring, snow/glacier studies and ingestion of historical digital sensor data in data cube framework for doing data analytics and generation of value added information. Currently, IRS-1C/1D georeferenced data products have system level location accuracy go up to 1 Km. In addition, the location error of IRS-1C/1D datasets are increased to many kilometers especially in hilly terrain scenes. Such higher location accuracy error makes legacy IRS-1C/1D data not directly usable for any kind of space borne applications. In this paper, an efficient approach is presented to improve the geo-location accuracy of IRS-1C/1D data by incorporating inflight geometric calibration exercises. The rigorous sensor model is fed with primary image, ancillary orbit and attitude data, digital elevation model (DEM) and a reference image to produce ortho-rectified product. The major challenge is to do automatic feature matching of IRS-1C/1D data with current RS-2A reference image database to achieve sub-pixel level geometric error. As datasets are more than two decades older, the number of outliers in matched feature points are more. Also geo-location error of 1C/1D is varying a lot which makes it difficult to set the search window radius for feature matching. To handle this scenario, mode seeking based pruning mechanism is added with robust SIFT feature matching technique which is independent of search window size. Additionally, SIFT parameters are also fine-tuned to get more stable feature points. The approach developed is tested in many multi temporal IRS-1C/1D LISS-3 datasets. The geo-location error of the products generated is at sub-pixel level accuracy with respect to the reference. The IRS-1C/1D data products generated with this new updated chain can be used directly as analysis ready datasets.

Monte-Carlo Siamese Policy on Actor for Satellite Image Super Resolution

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2020

In the past few years supervised and adversarial learning have been widely adopted in various com... more In the past few years supervised and adversarial learning have been widely adopted in various complex computer vision tasks. It seems natural to wonder whether another branch of artificial intelligence, commonly known as Reinforcement Learning (RL) can benefit such complex vision tasks. In this study, we explore the plausible usage of RL in super resolution of remote sensing imagery. Guided by recent advances in super resolution, we propose a theoretical framework that leverages the benefits of supervised and reinforcement learning. We argue that a straightforward implementation of RL is not adequate to address illposed super resolution as the action variables are not fully known. To tackle this issue, we propose to parameterize action variables by matrices, and train our policy network using Monte-Carlo sampling. We study the implications of parametric action space in a model-free environment from theoretical and empirical perspective. Furthermore, we analyze the quantitative and qualitative results on both remote sensing and non-remote sensing datasets. Based on our experiments, we report considerable improvement over stateof-the-art methods by encapsulating supervised models in a reinforcement learning framework.

Download

Mars Surface Multi-decadal Change Detection Using ISRO’s Mars Color Camera (MCC) and Viking Orbiter Images

Communications in Computer and Information Science, 2021

Mars is a dynamic and active planet in our solar system, which attracts humans due to different g... more Mars is a dynamic and active planet in our solar system, which attracts humans due to different geological events continuously reshaping its surface. ISRO's Mars Color Camera (MCC) onboard Mangalyaan spacecraft send more than thousand images of planet Mars at varying spatial resolution, which is of utmost importance for doing surface change detection over Mars. In this paper, we have described a methodology for automated change detection using MCC and Viking images having more than decade separation in image acquisition. The processing steps includes geometric transformation of Viking color image to the same size as MCC, image registration using SIFT based feature matching technique and automated change detection using multi-variate alteration detection (MAD). The workflow chain developed is tested in bitemporal images from MCC and Viking having more than 20 years' time span difference covering Esylum and Amentia quadrangles of Mars. The result shows the change detection map generated using MCC and Viking images, which focus the changing landscape of Mars due to wind streaks, dust deposits, landslides, lava flows and new impact craters formation.

Download

Automatic Cloud Detection in Resourcesat LISS-3 Data using Spectral Angle Similarity and Mode Seeking Reference Spectra

2020 IEEE 15th International Conference on Industrial and Information Systems (ICIIS), 2020

Optical remote sensing images contain cloud pixels, which need to be detected and masked for diff... more Optical remote sensing images contain cloud pixels, which need to be detected and masked for different earth observation studies and applications. In literature, various algorithms are reported which can detect cloud in multi-spectral remote sensing images. In this paper, we have developed an approach based on spectral angle computation that measures the radiometric gap between a reference pixel and input image pixels. Reference spectra is determined using mode seeking technique by taking multiple cloud pixels from multi-temporal images covering different land terrain. The processing chain is developed and tested in Resourcesat LISS-3 surface reflectance data. Result section shows that technique developed can determine thick cloud pixels with high confidence interval and compared with other novel cloud detection methods.

EPOCH: enhanced procedure for operational change detection using historical invariant features and PCA guided multivariate statistical technique

Geocarto International, 2022

FIRM: Framework for Image Registration Using Multistage Feature Detection and Mode-Guided Motion Smoothness Keypoint Optimization

IEEE Transactions on Geoscience and Remote Sensing, 2021

Remote-sensing image registration is a pivotal preprocessing step for earth observation data anal... more Remote-sensing image registration is a pivotal preprocessing step for earth observation data analytics. In this regard, georeferencing corrects the systematic geometric degradation in the image. However, it is difficult to achieve subpixel geometric accuracy across multitemporal scenes. This article focuses on hindmost part of geometric correction that uses reference layer and feature detection in hierarchical stages to improve the geometric fidelity of images at subpixel level. The methodology developed is based on patch affine-oriented fast and brief with mode-guided tiled scale invariant feature transform (MT-SIFT) techniques in a coordinate manner at a multistage processing architecture, which we refer to as FIRM. Motion smoothness constraint (MSC) keypoint correspondence optimization is used in FIRM to remove the outliers at gross stage and estimate segmented affine transformation parameters at finer stage. The automatic coregistration pipeline is evaluated in Indian Resourcesat multispectral camera images covering diverse landscapes. The capability of the designed framework is demonstrated to handle relatively large geometrical error. With more than a decade difference in acquisitions, multitemporal images are superimposed over each other and compared with state-of-the-art feature-based methods. The potential of the proposed approach FIRM is assessed on multisatellite imagery acquired from Resourcesat-2 and Landsat −8. It is observed that the root mean square error (RMSE) between coregistered images is 0.12 pixel at a spatial resolution of 5 m.

Automatic image registration framework for Remote Sensing Data using Harris Corner Detection and Random Sample Consensus (RANSAC) Model

Image registration is a fundamental image processing task to match and align physically two image... more Image registration is a fundamental image processing task to match and align physically two images which could have been imaged by different sensors, view angles or and at different times. It is necessary to have robust single frame image registration software especially an automated one. Automatic image registration framework overlays two images for geometric conformity aligning common features by establishing a transformation model using distinguishable feature points collected simultaneously in both the images in a completely un assisted manner. The critical steps in image registration are collection of feature points and estimating a spatial transformation especially when outliers are present besides feature matching and resampling the slave image to the master image geometry. In this paper, the details and merit of employing automatic Harris corner detection and building a transformation model using Random Sample Consensus (RANSAC) algorithm is brought out while registering a p...

S2A: Wasserstein GAN with Spatio-Spectral Laplacian Attention for Multi-Spectral Band Synthesis

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2020

Intersection of adversarial learning and satellite image processing is an emerging field in remot... more Intersection of adversarial learning and satellite image processing is an emerging field in remote sensing. In this study, we intend to address synthesis of high resolution multi-spectral satellite imagery using adversarial learning. Guided by the discovery of attention mechanism, we regulate the process of band synthesis through spatio-spectral Laplacian attention. Further, we use Wasserstein GAN with gradient penalty norm to improve training and stability of adversarial learning. In this regard, we introduce a new cost function for the discriminator based on spatial attention and domain adaptation loss. We critically analyze the qualitative and quantitative results compared with stateof-the-art methods using widely adopted evaluation metrics. Our experiments on datasets of three different sensors, namely LISS-3, LISS-4, and WorldView-2 show that attention learning performs favorably against state-of-theart methods. Using the proposed method we provide an additional data product in consistent with existing high resolution bands. Furthermore, we synthesize over 4000 high resolution scenes covering various terrains to analyze scientific fidelity. At the end, we demonstrate plausible large scale real world applications of the synthesized band.

Download

An Efficient Machine Learning Approach for Atmospheric Correction

Journal of the Indian Society of Remote Sensing, 2021

The earth observation using remote sensing images is an inquisitive way to explore and evaluate t... more The earth observation using remote sensing images is an inquisitive way to explore and evaluate the geo-resources of any specific area on the globe. In this regard, Indian Resourcesat-2A (R2A) remote sensing satellite plays an important role in monitoring the critical resources of our planet using its unique three tier imaging mechanism. Optical sensors on-board R2A have good spatial temporal resolution for diverse space borne applications. Most of these applications requires Surface Reflectance (SR) data product by removing the effects of intermittent atmospheric scattering and absorption. Radiative Transfer Models (RTM) are used to perform atmospheric correction which are computationally intensive, thus a look-up table (LUT) is utilized to interpolate intermediate values as a trade-off between accuracy and speed. However, the process of interpolation too becomes very compute intensive when a large enough LUT is used. The paper provides an approach to remove this trade-off by using multi-layered deep network to model interpolation as a regression problem. The proposed method generates highly accurate Deep SR product with a significant reduction in turn-around time. The experimental result shows that a speedup of 5x is achieved with the developed framework as compared to conventional interpolation-based approach for generation of R2A LISS-3 Deep SR scene data product. The Deep SR product is compared with pure 6SV generated product and R2 value found to be 0.97 (Green), 0.97 (Red), 0.98 (NIR) and 0.98 (SWIR), respectively. To check the efficacy of the framework, the LISS-3 Deep SR product is also compared with closest acquisition Landsat-8 SR product and ground truth values obtained through vicarious calibration. The maximum relative deviation error found to be 1.34%, 1.82%, 3.25% and 2.16% for Green, Red, NIR and SWIR channels, respectively.

Co-Registration of LISS-4 Multispectral Band Data Using Mutual Information-Based Stochastic Gradient Descent Optimization

Current Science, 2017

We propose a solution for automatic co-registration of LISS-4 MX radiometrically conditioned mult... more We propose a solution for automatic co-registration of LISS-4 MX radiometrically conditioned multispectral images issue by considering an optimization problem in which mutual information-based approach is used. Co-registration of multispectral images from the same sensor may also be a tough problem to tackle, when the payload imaging geometry is complex. The multispectral images acquired by ISRO Resourcesat-1/2 LISS-4 MX class of sensors pose such problems and demand an automatic registration solution for systemcorrected product generation to cater to user needs. Optical remote sensing image registration is assisted by image geo-referencing or navigation information along with components such as feature detection, matching, correspondence, and resampling the input image to the reference geometry. Intensity-based methods employ an iterative registration framework, where similarity metric based image matching and correspondence is refined to find out optimum transform parameters. We could successfully employ mutual information-based adaptive stochastic gradient descent optimization algorithm to do sub-pixel level satellite image registration tasks by a careful choice of parameters and models related to metric, transform, optimizer, and interpolator in a robust image registration framework which is automatic for different terrain data. The performance is also compared to a recent scale invariant feature transform (SIFT)based registration method.

Download

Automatic Registration of INSAT-3D Daily Images Using Mutual Information and Stochastic Optimization Technique

Journal of the Indian Society of Remote Sensing, 2018

An automatic image registration approach is presented here can be used to register daily images o... more An automatic image registration approach is presented here can be used to register daily images of Indian geostationary satellite system INSAT-3D acquired every 30 min' interval without use of any ground control points (GCPs). There is always a pressing need to register meteorological images that are acquired over earth from geostationary platforms every 15-30 min, covering almost one-third of the earth. Weather forecast activities include derivation of atmospheric motion vectors, which demand immediate processing of such images to a reasonable accuracy in terms of its relative location accuracy. Generally followed approaches make use of image navigation models and GCPs drawn from known landmarks in land ocean boundaries and correlate image features before estimating a transform to warp the current acquisition to a known geometry. However, the hierarchical (coarse to fine) approach explained here makes use of intensity based Mutual Information as a similarity measure from a population of pixels selected randomly and uses stochastic gradient descent optimizer to estimate an affine transform between registering image pair, delivers satisfactory results.

Download

An Automated Software Framework for Evaluation and Improvement of Absolute Geometric Location Accuracy of Multispectral Remote Sensing Satellite Imagery

ISPRS - International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2018

Geometric accuracy is an important parameter for quality assessment of a data product and is vita... more Geometric accuracy is an important parameter for quality assessment of a data product and is vital for certain applications aiming on improving and bringing precision in data products. Automatic geometric accuracy evaluation of the satellite image is attempted by matching a known, surveyed location typically a ground control point (GCP) calibrated using a differential global positioning system (DGPS), verified using Google Maps, to the corresponding identifiable feature in an image product. The requirement for this development is to address the nonuniformity in the available data products in terms of coordinates reference system, resolution and available bands, which the software overcomes successfully by benefiting from the classes and functions available in openly available GIS libraries. RMSE of 0.8 pixels is found in analysis for the chosen data. Further, an algorithm is worked up to rectify the image for this geometric shift.

Download

Minnaert Topographic Correction of Mars Color Camera Images from Mars Orbiter Mission

2015 International Conference on Information Technology (ICIT), 2015

Mars Color Camera (MCC) images obtained from Mars Orbiter Mission (MOM) are gaining scientific po... more Mars Color Camera (MCC) images obtained from Mars Orbiter Mission (MOM) are gaining scientific popularity since MOM insertion into an elliptical orbit around Mars on 24th Sep, 2014. Planetary remotely sensed images are corrected for topographic effects to normalize the radiance measures before considering the data for science analysis. It is proposed here to use non Lambertian Minnaert semi empirical approach for correcting MCC images that are used for deriving results for Mars surface science. The methodology outlined here uses terrain parameters such as slope and aspect derived from Mars Orbiter Laser Altimeter (MOLA) digital elevation model (DEM). Topographically corrected images were evaluated for the improvement in its radiometry quantitatively and it is found to reduce topographic shading and improve the image quality.

debajyoti dhar

Uploads

Papers by debajyoti dhar

Log In