classSel

Description

classSel is a utility designed to extract and analyze backscatter characteristics from angular look-up tables (LUTs) of multibeam data. It calculates various statistical parameters, such as mean backscatter, slope of the angular response, and specific values at nadir, center, and tail angles, within user-defined spatial windows.

The tool operates on pre-generated LUT files (.lut and .num), which store backscatter values and associated metadata (profile, beam, grazing angle, intensity) for individual soundings. It effectively performs a moving window analysis over these LUTs to classify or characterize the seafloor.

Usage 

classSel -lut <LUThandle> [-step <pixels>] [-v]

Arguments

Option Description  
-lut <LUThandle> Required. The handle (base name) for the Look-Up Table files. The tool expects <LUThandle>.lut and <LUThandle>.num files.  
-step <pixels> Specifies the size of the square analysis window (in pixels) for both X and Y dimensions. 10
-v Enable verbose output.  

How It Works

  1. Load Information (load_up_info):
    • Opens the .lut (Look-Up Table) and .num (number of samples) files using the provided LUThandle.
    • Reads the JHC headers from both files to get dimensions (dx, dy) and data_type.
    • Reads the num array (containing the count of valid samples per pixel) into memory.
    • Note: The .lut file (containing Sb_value structures) is not fully loaded into memory but accessed via fseek for individual pixels due to its potentially large size.
  2. Windowed Analysis:
    • The tool iterates through the num_header.dy and num_header.dx dimensions, effectively moving a square analysis window of size step by step pixels.
    • For each window:
      • Calculate Angular Dependence (calc_ang_dep):
        • Resets Sbmean (sum of linear backscatter) and Sb_count for 90-degree bins (0-89 degrees).
        • Iterates through all pixels (k, i) within the current window.
        • For each pixel, it fseeks to the corresponding location in the .lut file and reads tot (num of samples) Sb_value structures.
        • For each Sb_value, it extracts the Sb (backscatter intensity, converted to linear scale) and angle (grazing angle).
        • Accumulates Sb in Sbmean[(int)angle + 0.5] and Sb_count[(int)angle + 0.5]++.
        • After processing all samples in the window, it calculates Sbmean[i] /= Sb_count[i] for each bin where data exists.
        • It then calculates Nadir, Centre, and Tail backscatter values (in linear units, then converted to dB) by averaging Sbmean within specific angle ranges (80-90 deg for Nadir, 40-50 deg for Centre, 15-25 deg for Tail).
        • Least-Squares Fit (Commented Out): There’s commented-out code for performing a linear least-squares fit on the 10.0*log10(Sbmean[i]) vs. angle to determine Slope and Average. This is not currently active.
  3. Output Grids:
    • After analyzing all windows, it creates four output .r4 (unsigned char) files:
      • <LUThandle>.mea: Mean backscatter (Centre value).
      • <LUThandle>.slp: Slope of the angular response (currently Slope is not actively calculated in calc_ang_dep).
      • <LUThandle>.nad: Nadir backscatter.
      • <LUThandle>.tai: Tail backscatter.
    • These output grids are created with the same header as the num file, and each pixel represents the calculated parameter for the corresponding analysis window. The values are scaled to 0-255 based on dB_min and dB_max.
  4. Memory Management: Allocates and frees memory for the output grids.

Output Files

  • <LUThandle>.mea: JHC-format .r4 grid of mean backscatter.
  • <LUThandle>.slp: JHC-format .r4 grid of slope of angular response.
  • <LUThandle>.nad: JHC-format .r4 grid of nadir backscatter.
  • <LUThandle>.tai: JHC-format .r4 grid of tail backscatter.

Dependencies

  • OMG_HDCS_jversion.h: For JHC_header structure.
  • support.h: For general utility functions and error handling.

Notes

This tool is used for seafloor classification by deriving statistical properties of backscatter and grazing angle within spatial windows. The output grids can then be used in further classification algorithms or for visualizing seabed texture. The current implementation uses 8-bit output for the .r4 grids, which implies a scaling of the float values to fit the 0-255 range.