T13_Accuracy_Assessment_Part1_3slides

T13_Accuracy_Assessment_Part1_3slides - Geography 333...

Info iconThis preview shows pages 1–6. Sign up to view the full content.

View Full Document Right Arrow Icon
1 Geography 333 Remote Sensing I Topic 13: Accuracy Assessment 2 Readings Topic 13: Chapter 13 – Thematic Map Accuracy Assessment Topic 14: Chapter 12 – Change Detection Topic 15 (?): (i) Microwave (ii) Lidar, (iii) Thermal Imagery – HEAT project 3 Outline Training Aids/ Feature Selection Histograms Scatter plots Divergence tools Confusion Matrices and Statistics Sampling Map Lineage and Metadata
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
2 4 Classification overview Define the classification problem: Determine the study area boundary Identify classes of interest Acquire the appropriate data: Remote sensing, DEM, ancillary data ‘Ground truth’ information of study area Perform pre-processing tasks: Radiometric correction Geometric correction Data integration Perform image processing for thematic information extraction: Select classification logic and algorithms Perform image transformation / parameter selection Extract initial training data (if required) Select most appropriate input variables Extract final training data (if required) Extract thematic information Assess classification accuracy Repeat until satisfied with accuracy results Report and distribute Image and map lineage report Metadata Distribute digital/analog products 5 Thematic Information Extraction (details) Fig 9.1 – Jensen – pg 339 6 Training Aids/Feature Selection Goal : evaluate the spectral class separability and provide guidance for the iterative process of training set refinement Graphical plots of training data Statistical measures of separability
Background image of page 2
3 7 Training Stage Goal : to assemble a set of statistical measures that accurately characterize the spectral patterns of the desired information classes Must be both representative and complete All spectral classes constituting each information class must be fully represented 8 Training Rules Minimum sample size for training areas is 10n to 100n, where n=number of input channels, this value is in pixels, not in objects Better to get several representative samples from several sites throughout the scene, rather than one big sample Training data (and input datasets) MAKES the classification Training refinement is an important part of the process 9 Histogram Check the range and distribution of the data in each class, and know the statistical requirements of the classifier. In a MLC: is it normally distributed?
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
4 10 Graphical Training Tools: Histograms 11 Graphical Training Tools: Scatterplots 12 Scatter plots with image dancing (ENVI) Load ROI and evaluate how compact its ‘signature’ is in the scatterplot of two channels. Look for overlap.
Background image of page 4
5 13 Statistical Training Tools: Divergence Divergence* : a relatively simplistic measure of class separability Transformed divergence accounts for the over representation of outlier classes Values over 1900* = good under 1700* = bad * In PCI, divergence statistics are scaled to 0-2.0 14 15 Compute ROI Separability (in Envi) This option computes the spectral separability between selected ROI pairs for a given input file. Both the J
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 6
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 14

T13_Accuracy_Assessment_Part1_3slides - Geography 333...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online