Work Package 5: Signal Processing and Seafloor Classification Jarosław Tęgowski and Karolina Trzcińska Marine Geophysics Lab. Institute of Oceanography University of Gdańsk, Poland Kick-Off Meeting, Kiel, 5-6 December 2017
GENERAL CONCEPT OF ECOMAP WITH USED METHODS
WP 5 Objectives: Efficient processing and reliable algorithms for extraction of highly dimensional datasets, Generation of habitat specific indicator variables, Validation of calibrated echosounders and algorithms at reference test sites. Task 5.1: Spectral, wavelet, and statistical analyses of geomorphology of Digital Elevation Model (DEM) constructed on basis of opto-acoustics (M1-M17) Task 5.2: Model consultation and validation of calibrated sounders and LiDAR classifiers (M17-M24) Task 5.3: Signal processing methods of optical and acoustical data to extract additional information from the seabed surface, from volume backscattering of the shallow subbottom, and from backscatter signals in the water column by vegetation (M7)
ACOUSTIC AND LIDAR CLASSIFICATION OF BOTTOM SEDIMENTS AND MORPHOLOGIC FORMS HABITATS IDEAS OWN ALGORITHMS Single beam echosounder Multi beam echosounder Side scan sonar Parametric echosounder, subbottom profiler Lidar.. Acoustic underwater positioning - USBL Multibeam echosounder Singlebeam echosounder Vibro sounder Sidescan sonar and magnetometer Remotely operated underwater vehicle - ROV Satellite positioning system GPS (< 2 cm) Environmental monitoring buoy Acoustic subbottom profiler
RoxAnn QTC BioSonics Inc... SINGLE BEAM ECHOSOUNDER Seafloor classification systems
Acoustic classification of sediments using SBE Example North Sea
GENERAL SCHEME FOR THE ACOUSTIC CLASSIFICATION OF SEDIMENTS 83 parameters of echo envelope Fuzzy logic + k-means, PCA, 150kHz Fuzzy logic + k-means, PCA, 66kHz Reduction of the number of parameters Principal Component Analysis Factor Analysis Fuzzy logic + k-means, FA, 150kHz Fuzzy logic + k-means, FA, 66kHz Classification algorithms
Angular dependency of backscattered intensity seafloor classification medium grained sand Bottom backscattered intensity contais information about: - type of sediment - geomorfologic forms fine sand mud
PARAMETRIZATION OF ANGULAR DEPENDENCY OF BACKSCATTERED INTENSITY - SEAFLOOR CLASSIFICATION Wavelet transform parameters Spectral prameters Statistical parameters Fractal parameters
Map of backscattered signal intensity Rewal area 1 ridges boulder clay and gyttja; 2 fine and medium grained sand; isolated outcrops of lacustrine gyttja and mud covered with a layer of sand (up to 30cm); 3 slopes of elevations, medium-grained sands
CLASSIFICATION ALGORITHM I Multibeam echosounder echo intensity 19 parameters of Fourier transformation computed for angular dependency of backscatter intensities Classified bottom sediments Segmentation algorithm Unsupervised Kohonen s neural network
CLASSIFICATION ALGORITHM II Multibeam echosounder echo intensity 9 parameters computed for angular dependency of backscatter intensities energies of wavelet transformation (Coiflet wavelets) Classified seafloor geomorphological forms Segmentation algorithm Unsupervised Kohonen s neural network
COMPARISON OF CLASSIFICATION RESULTS seafloor geomorphologic forms types of sediment
MULTIBEAM ECHOSOUNDER SNIPPETS IDEA ONGOING WORK ICES COOPERATIVE RESEARCH REPORT NO. 286
APL-UW* MODEL OF BOTOOM BACKSCATTERING STRENGTHS at 100 khz BBS *Applied Physics Lab. Univ. of Washington
SNIPPETS NORBIT S MULTIBEAM ECHOSOUNDER IDEA OF CLASSIFICATION ALGORITHM Calibrated echosounder Real levels of the Bottom Backscattering Strength for different angles of incidence Shape parameters of snippets The above parameters will be input to the classification algorithm
Object-Based Image Analysis marine seafloor mapping Łukasz Janowski PhD student General Workflow: 1.Processing of the MBES DEM and backscatter mosaic 2.Feature extraction and selection 3.OBIA segmentation and classification, accuracy assessment Trimble ecognition software Standard deviation of backscatter, Gray Level Co-occurance Matrix (GLCM), textural features ArcGIS - Bethic Terrain Modeler Toolbox Rugosity, slope, variance, aspect, northness, eastness, curvature, profile curvature, planar curvature, BPI (Bathymetric Position Index), sd (standard deviation) of some features CART Classififcation and Regression Trees allow to distinguish characteristic features of the image (eg. DEM and BBS)
OBIA segmentation and classification, accuracy assessment Łukasz Janowski Multiresolution segmentation of scale 280 and CART classifier Result: set of if-then rules dependent of backscatter and rugosity Error matrix Reference Class User MS C FS VFS Sum MS 2 0 0 1 3 C 0 0 0 0 0 FS 0 0 2 0 2 VFS 0 0 0 3 3 Sum 2 0 2 4 (muddy sand) (clay) (fine sand) (very fine sand) Producer 1 undefined 1 0.75 User 0.666667 undefined 1 1 Overall Accuracy 0.875 KIA 0.809524
SIDESCAN SONAR Seafloor classification systems - IDEAS Object-Based Image Analysis same metod as presented for MBES 2D Wavelet Transform parameters EXAMPLE OF MULTIPLE LEVEL WAVELET DECOMPOSITION APPLIED FOR SIDESCAN SONAR IMAGERY strong reflection stony bottom mound acoustical shadow vegetations
( ) ( ) ( ) D N V N H N N D N V N H N D V H D D D A D D D D D D S,,,,,...,, 1 1 1 1 1 1 + + + = MULTIPLE LEVEL WAVELET DECOMPOSITION TREE Biorthogonal wavelets - Bior3.7, Bior6.8
WAVELET DECOMPOSITION AND SYNTHESIS OF SIDESCAN SONAR IMAGERY DECOMPOSITION Multiple Level Wavelet Decomposition Approximation A2 Horizontal Detail H3 Diagonal Detail D3 200 400 600 800 1000 1200 1400 50 100 150 200 250 300 350 400 450 500 200 400 600 800 1000 1200 1400 50 100 150 200 250 300 350 400 450 500 200 400 600 800 1000 1200 1400 50 100 150 200 250 300 350 400 450 500 SYNTHESIS Segmentation algorithms Unsupervised Kohonen s neural network or Fuzzy c-means cluster analysis
CLASSIFICATION OF BOTTOM STRUCTURES USING PARAMETRICAL ECHOSOUNDER AND SUBBOTTOM PROFILER DATA Extraction of Seabed/Subsurface Features in a Potential CO2 Sequestration Site in the Southern Baltic Sea, Using Wavelet Transform of High-resolution Sub-Bottom Profiler Data 4 clusters 651600000 651600000 651500000 651500000 Northing [m] Northing [m] 3 clusters 651400000 651300000 651400000 651300000 651200000 651200000 651100000 651100000 615000000 615200000 Easting [m] 615400000 615000000 615200000 Easting [m] 615400000
LIDAR data 1.Object-Based Image Analysis same method as used for MBES data 2.Statistical and Spectral Features of Corrugated Seafloor 2D Fourier Analysis Example of autocorrelation length computation using 2D FFT and the Wiener Khinchin theorem MBES data Hornsund fjord, West Spitsbergen Autocorrelation length DEM - seafloor
WP 5 Strengths, Weaknesses, Opportunities, and Threats (SWOT) Strengths We are experienced in acoustical classification of bottom sediments, geomorphological forms and structures. We have good quality equipment. Weaknesses Too many methods, too many different kinds of data, too many algorithms. Opportunities We have an opportunity to make progress in methods of non-invasive seabed investigation. Threats Too short time for testing and improvement of all new algorithms.
THANK YOU VERY MUCH FOR YOUR ATTENTION