In this lab for REM-617-01: Image Analysis & Information Extraction, we performed image manipulation and atmospheric correction using a Landsat ETM+ scene in ENVI. The assignment required stacking spectral bands, interpreting RGB composites, creating a Region of Interest (ROI), building a mask, computing statistics, and applying Dark Object Subtraction (DOS). The objective was to move from raw satellite data to a cleaner, analysis-ready image.
Stacking the Landsat Bands
The first task was stacking seven individual Landsat band files into one multi-band image. Each file represented a different spectral band, and stacking allowed ENVI to treat them as one dataset. Spectral analysis depends on comparing reflectance values across multiple wavelengths, so this step is foundational.
After stacking, the image appeared in the Available Bands List as a single layered file, confirming that all seven bands were successfully merged.
Interpreting RGB 742 and RGB 321
We created two color composites:
- RGB 742
- RGB 321
RGB 321 resembles natural color imagery. Vegetation appears green, water appears dark, and built-up areas appear gray or brown.
RGB 742 enhances vegetation because Band 4 (near infrared) reflects strongly from plant material. Vegetation appears bright red in this composite, while water remains very dark. Linking both displays allowed direct comparison of how features respond differently across spectral combinations.
This comparison demonstrated how band combinations influence interpretation and land cover identification.
Creating the ROI and Building the Mask
The Landsat scene included a black background surrounding the valid image area. These background pixels had values of zero and would distort statistical calculations if included.
To correct this, we created a Region of Interest (ROI) covering the valid image area. Initially, the ROI I created was not large enough and did not fully capture the usable image. After reviewing pixel values near the edges, I increased the ROI size to ensure complete coverage of valid pixels. This adjustment was necessary to maintain statistical accuracy.
The ROI was then used to build a mask image.
In the mask:
- Pixels inside the ROI had a value of 1
- Pixels outside had a value of 0
This means:
- 1 = Include pixel in analysis
- 0 = Exclude pixel
The mask functioned as a filter, ensuring that statistical calculations were performed only on valid image pixels.
Computing Statistics Using the Mask
Using the mask, we computed basic statistics and histograms for all seven bands. Because the ROI was corrected to fully cover the valid image area, the statistics accurately represented the scene without background contamination.
The spectral trend showed:
- A strong peak in Band 4 (near infrared), consistent with vegetation reflectance.
- Elevated values in Band 6, the thermal band.
- Lower reflectance in visible bands depending on surface materials.
The statistics confirmed that:
- The ROI adjustment ensured complete data coverage.
- The mask properly excluded background pixels.
- The spectral response aligned with expected remote sensing behavior.
Atmospheric Correction Using Dark Object Subtraction (DOS)
The final step was atmospheric correction using the Dark Object Subtraction method.
We subtracted the minimum value of each band derived from the masked statistics:
| Band | Minimum Value Subtracted |
|---|---|
| 1 | 67 |
| 2 | 29 |
| 3 | 23 |
| 4 | 94 |
| 5 | 55 |
| 6 | 132 |
| 7 | 17 |
The DOS method assumes that the darkest object in the image should have near-zero reflectance. Any measured value above zero is treated as atmospheric noise. Subtracting these values reduced atmospheric effects and produced a relatively corrected image.
After correction, new statistics were computed using the same mask to ensure consistency.
The corrected image showed reduced baseline values while maintaining spectral relationships across bands.
Conclusion
This lab demonstrated that raw satellite imagery is not analysis-ready. It must be stacked, interpreted, filtered, and corrected before meaningful analysis can occur.
The adjustment of the ROI highlighted that image processing is iterative. Small oversights, such as incomplete ROI coverage, can affect statistical integrity. Careful validation and correction are part of responsible remote sensing analysis.
From stacking to masking to atmospheric correction, each step improved the quality and reliability of the dataset for future analysis.