Impact of SAR data filtering on crop classification accuracy

Slides:



Advertisements
Similar presentations
Beyond Spectral and Spatial data: Exploring other domains of information GEOG3010 Remote Sensing and Image Processing Lewis RSU.
Advertisements

CSCE 643 Computer Vision: Template Matching, Image Pyramids and Denoising Jinxiang Chai.
On Estimation of Soil Moisture & Snow Properties with SAR Jiancheng Shi Institute for Computational Earth System Science University of California, Santa.
Sar polarimetric data analysis for identification of ships S. Swarajya lakshmi ADRIN, Dept. of Space, Govt. of India India Geospatial Forum – 14 th International.
Mathematical Modelling in Geography: II GEOG2021 Environmental Remote Sensing.
Multiple Criteria for Evaluating Land Cover Classification Algorithms Summary of a paper by R.S. DeFries and Jonathan Cheung-Wai Chan April, 2000 Remote.
Detecting Image Region Duplication Using SIFT Features March 16, ICASSP 2010 Dallas, TX Xunyu Pan and Siwei Lyu Computer Science Department University.
A REAL-TIME VIDEO OBJECT SEGMENTATION ALGORITHM BASED ON CHANGE DETECTION AND BACKGROUND UPDATING 楊靜杰 95/5/18.
Despeckle Filtering in Medical Ultrasound Imaging
Troy P. Kling Mentors: Dr. Maxim Neumann, Dr. Razi Ahmed
MULTITEMP 2005 – Biloxi, Mississippi, USA, May 16-18, 2005 Remote Sensing Laboratory Dept. of Information and Communication Technology University of Trento.
Department of Biophysical and Electronic Engineering (DIBE)- Università di Genova- ITALY QUALITY ASSESSMENT OF DESPECKLED SAR IMAGES Elena Angiati, Silvana.
Competence Centre on Information Extraction and Image Understanding for Earth Observation Matteo Soccorsi (1) and Mihai Datcu (1,2) A Complex GMRF for.
LAND-USE MAPPING USING COARSE RESOLUTION SAR DATA AT THE OBJECT LEVEL EXPLOITING ANCILLARY OPTICAL DATA ALDRIGHI M., GAMBA P. Remote Sensing Team Università.
1 Vladimir Lukin 31/08/2009 Processing of Images Based on Blind Evaluation of Noise Type and Characteristics Processing of Images Based on Blind Evaluation.
ASPRS Annual Conference 2005, Baltimore, March Utilizing Multi-Resolution Image data vs. Pansharpened Image data for Change Detection V. Vijayaraj,
Ping Zhang, Zhen Li,Jianmin Zhou, Quan Chen, Bangsen Tian
National Aerospace University “Kharkov Aviation Institute” SPIE Remote Sensing Performance prediction for 3D filtering of multichannel images Oleksii.
Fuzzy Entropy based feature selection for classification of hyperspectral data Mahesh Pal Department of Civil Engineering National Institute of Technology.
1 Lecture 1 1 Image Processing Eng. Ahmed H. Abo absa
1 Vladimir Lukin 19/08/2009 Lossy Compression of Images Corrupted by Mixed Poisson and Additive Gaussian Noise Vladimir V. Lukin a, Sergey S. Krivenko.
Authors: Sriram Ganapathy, Samuel Thomas, and Hynek Hermansky Temporal envelope compensation for robust phoneme recognition using modulation spectrum.
Paddy Damage Assessment due to Cyclonic Storm using Remotely Sensed Data By ABHIJAT ARUN ABHYANKAR October 4, 2010.
ESTUARY WETLAND DETECTION IN SAR IMAGES Presented By Yu-Chang Tzeng.
FSU Jena – Department of Earth Observation CREATION OF LARGE AREA FOREST BIOMASS MAPS FOR NE CHINA USING ERS-1/2 TANDEM COHERENCE Oliver Cartus (1), Christiane.
Advances in digital image compression techniques Guojun Lu, Computer Communications, Vol. 16, No. 4, Apr, 1993, pp
Computer-based identification and tracking of Antarctic icebergs in SAR images Department of Geography, University of Sheffield, 2004 Computer-based identification.
1 IMPROVED NOISE PARAMETER ESTIMATION AND FILTERING OF MM-BAND SLAR IMAGES Vladimir V. Lukin, Nikolay N. Ponomarenko, Sergey K. Abramov Dept of Transmitters,
On Estimation of Soil Moisture with SAR Jiancheng Shi ICESS University of California, Santa Barbara.
Autonomous Robots Vision © Manfred Huber 2014.
BOT / GEOG / GEOL 4111 / Field data collection Visiting and characterizing representative sites Used for classification (training data), information.
Kussul Nataliia, Shelestov Andrii, Skakun Sergii Space Research Institute of NAS of Ukraine and SSA of Ukraine Kyiv National University of Environmental.
Face Image-Based Gender Recognition Using Complex-Valued Neural Network Instructor :Dr. Dong-Chul Kim Indrani Gorripati.
0 Riparian Zone Health Project Agriculture and Agri-Food Canada Grant S. Wiseman, BS.c, MSc. World Congress of Agroforestry Nairobi, Kenya August 23-28,
Classification Method Validation for Rice Mapping Using ENVISAT APS Data Erxue CHEN(1), Zengyuan LI(1), BingxiangTan(1) , Wei He(1), Bingbai LI(2) (1)Institute.
Chapter 8 Lossy Compression Algorithms. Fundamentals of Multimedia, Chapter Introduction Lossless compression algorithms do not deliver compression.
1 Marco Carli VPQM /01/2007 ON BETWEEN-COEFFICIENT CONTRAST MASKING OF DCT BASIS FUNCTIONS Nikolay Ponomarenko (*), Flavia Silvestri(**), Karen.
Presented by: Kumar Magi. ( 2MM07EC016 ). Contents Introduction Definition Sensor & Its Evolution Sensor Principle Multi Sensor Fusion & Integration Application.
Anthropogenic Change Detection in Alberta Anthropogenic Change Detection in Alberta: A Semi-automated Extraction Technique from SPOT5 Panchromatic Satellite.
A School of Mechanical Engineering, Hebei University of Technology, Tianjin , China Research on Removing Shadow in Workpiece Image Based on Homomorphic.
National Aerospace University of Ukraine IS&T/SPIE Electronic Imaging METRIC PERFORMANCE IN SIMILAR BLOCKS SEARCH AND THEIR USE IN COLLABORATIVE.
An improved SVD-based watermarking scheme using human visual characteristics Chih-Chin Lai Department of Electrical Engineering, National University of.
Process and Content of Ukraine's Agricultural Land Use Monitoring
Chapter 8 Lossy Compression Algorithms
MAIN PROJECT IMAGE FUSION USING MATLAB
Nikolay N. Ponomarenkoa, Vladimir V. Lukina,
Aleksey S. Rubel1, Vladimir V. Lukin1,
VEGA-GEOGLAM Web-based GIS for crop monitoring and decision support in agriculture Evgeniya Elkina, Russian Space Research Institute The GEO-XIII Plenary.
A 2 veto for Continuous Wave Searches
PERFORMANCE ANALYSIS OF VISUALLY LOSSLESS IMAGE COMPRESSION
Scatter-plot Based Blind Estimation of Mixed Noise Parameters
Built-up Extraction from RISAT Data Using Segmentation Approach
Bag-of-Visual-Words Based Feature Extraction
Alexey Roenko1, Vladimir Lukin1, Sergey Abramov1, Igor Djurovic2
Compression for Synthetic Aperture Sonar Signals
Improving the Performance of Fingerprint Classification
Speech Enhancement with Binaural Cues Derived from a Priori Codebook
Outlier Processing via L1-Principal Subspaces
SPECKLE REDUCING FOR SENTINEL-1 SAR DATA
Prof. Nataliia Kussul, Space Research Institute NASU-SSAU
Efficient Estimation of Residual Trajectory Deviations from SAR data
Nikolay Ponomarenkoa, Vladimir Lukina, Oleg I. Ieremeieva,
Closson, Abou Karaki, al-Fugha
Vehicle Segmentation and Tracking in the Presence of Occlusions
Another Cambridge physicist... Bounces man-made radar waves off target
Computer Vision Lecture 16: Texture II
Use of Geospatial Data for SDG Monitoring
2011 International Geoscience & Remote Sensing Symposium
Review and Importance CS 111.
IPSN19 杨景
Presentation transcript:

Impact of SAR data filtering on crop classification accuracy UkrCon2017 1  Impact of SAR data filtering on crop classification accuracy  M. Lavreniuk1, N. Kussul1, M. Meretsky1, V. Lukin2, O. Rubel2, S. Abramov2 1Department of Space Information Technologies and Systems Space Research Institute NAS Ukraine and SSA Ukraine Kyiv, Ukraine 2Department of Transmitters, Receivers and Signal Processing, National Aerospace University, Kharkov, Ukraine

Radar remote sensing (image filtering) 2 Synthetic aperture radar (SAR) is a tool employed for various applications of remote sensing as: hydrology; ecology; forestry; agriculture. This is due to the following advantages of SAR as: appropriate spatial resolution; possibility to operate during day and night, in almost all weather conditions; availability of SAR data; periodicity of observations Radar images can be used: jointly with optical or infrared ones; Separately if obtained for several frequencies and/or polarizations (see Sentinel images →); Jointly if obtained in multi-temporal mode.

Radar remote sensing (speckle) 3 Drawback: Radar images acquired by SAR inevitably suffer from noise-like phenomenon called speckle. Speckle is a specific kind of noise that has multiplicative nature and is often spatially correlated. Speckle properties depend upon several factors including is this intensity or amplitude image, what is number of looks. Noise-free, amplitude and intensity images (single look)

Radar remote sensing (speckle) 4 Speckle presence in negative way influences all procedures of SAR image processing as: segmentation; lossless compression; classification; object and edge detection; texture detection and feature determination. Denoising or despeckling is used to improve quality of SAR RS images and improve performance of subsequent application (classification, compression, etc.): Denoising efficiency can be characterized in different ways. This can be done using conventional criteria as output MSE or PSNR. Meanwhile, filtering efficiency can be characterized indirectly, via criteria that describe classification accuracy. Our goals: - to analyze filtering efficiency by determining its impact on crop classification; to study such properties of Sentinel SAR images that are important for filtering; to compare performance of different filters.

8x8 normalized DCT spectra for several images Sentinel SAR images (main properties) 6 It has been established that speckle is pure multiplicative. Its relative variance is about 0.05. Cross correlation of VV and VH components is about 0,85. Speckle is spatially correlated. Spatial correlation properties are practically the same for all considered image fragments and for both polarization components. 8x8 normalized DCT spectra for several images Notes: small indices correspond to low spatial frequencies; Estimates have been obtained in homogeneous image fragments using robust techniques. Analysis clearly shows that speckle is spatially correlated.

Requirements to despeckling and analyzed filters 7 Main requirements to filters: to suppress speckle; to preserve edges and fine details; to preserve texture; to preserve mean in homogeneous regions and ratios for multichannel data. Analyzed filters: Filters available in ESA SNAP Toolbox The DCT-based filters and, particularly, the conventional DCT-based filter that have demonstrated high efficiency for a wide class of images and noise models. There are modifications for pure multiplicative noise. Filtering is fast enough. Images corrupted by multiplicative and other types of signal dependent noises can be processed applying homomorphic and/or variance stabilizing transformations to produce additive noise. Then, they can be processed by standard DCT-based filter, A block matching 3D (BM3D) filter is considered to be the most effective in AWGN suppression for component (grayscale) images where this filter also exploits DCT-based denoising as a basic mechanism of noise removal. Homomorphic transform is needed.

Classification approach 11 4. Geospatial analysis (data fusion from the classification maps and vector information) 3. Map filtration (voting and weighted voting approaches with division parcels into the fields) 2. Universal deep learning approach for time series classification at the regional level 1. No-data pixels restoration (clouds and shadows) using self-organized Kohonen maps

Validation: JECAM experiments 12 Ukrainian JECAM site Cropland mask Kyiv region

Image without filtering; Refined lee filter; C) DCTF; D) MultiLogDCTF. Classification maps 13 Image without filtering; Refined lee filter; C) DCTF; D) MultiLogDCTF.

Filter OA, % / Kappa Results 14 No filter 82.6 / 0.7 Boxcar 86.4 / 0.76 Frost 86.7 / 0.77 Gamma map 86 / 0.75 IDAN 86.2 / 0.76 Lee 86.5 / 0.76 Lee Sigma 87/ 0.77 Median 85.8 / 0.75 Refined Lee 87.4 / 0.78 DCTF 87.7 / 0.78 DCTF HT 87.5 / 0.78 MultiLogDCTF 88.7 / 0.8

Conclusions The overall accuracy without any filters is 82.6%; 15 The overall accuracy without any filters is 82.6%; Utilizing of different methods for speckle reducing results in better classification maps - accuracy ranges from 85.8% to 88.7%; The most accurate and useful filter from ESA SNAP toolbox for crop mapping task is the Refined Lee; The proposed DCT-based filtering approach has outperformed all available filters in ESA SNAP toolbox; Importantly, that this filter increases UA and PA for each class, excluding forest and bare land.

16 Thank you!