The issue you're encountering, where the points used for accuracy assessment in ERDAS Imagine do not
align with the correct land cover classes, could be related to several factors. Here are some common
causes and potential fixes:
### 1. **Georeferencing Misalignment**:
- **Issue**: If the georeferencing process was not done accurately, there may be a spatial
misalignment between the classified image and the reference points used for accuracy assessment.
- **Fix**:
- **Recheck your georeferencing**: Ensure that the Ground Control Points (GCPs) used in the
georeferencing process are accurate and that the image is properly aligned with your base map or
reference.
- Use tools like **RMSE** (Root Mean Square Error) to assess the accuracy of your georeferencing
and try to minimize the error to an acceptable level (typically below 1 pixel).
### 2. **Projection Issues**:
- **Issue**: If the projection or coordinate system of your image differs from that of the reference
data (like the training points or validation data), the points may appear shifted.
- **Fix**:
- Ensure both your image and the accuracy assessment points are in the **same coordinate
system**. You can check and reproject your image or vector files to ensure they match.
- In ERDAS Imagine, go to **Metadata** and check the projection information of both your image
and the reference data.
### 3. **Pixel Shift**:
- **Issue**: Even small errors in georeferencing or during the classification process could cause pixel
shifts, leading to incorrect points being classified.
- **Fix**:
- **Refine your georeferencing**: Use more accurate GCPs and ensure that the image is well-aligned
with reference data.
- **Snap to grid**: In some cases, snapping your points or the classified map to a grid that matches
your classification grid might help.
### 4. **Image Resampling**:
- **Issue**: If the image was resampled during the classification or georeferencing process, it may
have caused some displacement of the data.
- **Fix**:
- Use **nearest neighbor resampling** when performing classification to avoid altering pixel values.
- Check if your resampling technique (e.g., bilinear, cubic convolution) is introducing distortions and
consider rerunning the classification with appropriate settings.
### 5. **Inaccurate Training Data**:
- **Issue**: If your training data or reference data are not accurately representing the land cover
classes or were misaligned, the accuracy assessment will show discrepancies.
- **Fix**:
- Ensure that the training data points were created using accurately georeferenced and classified
images.
- Verify the reference points used for accuracy assessment to ensure they are correctly placed and
represent the right land cover type.
### 6. **Check Image Alignment in Multiple Scales**:
- **Issue**: Sometimes the misalignment may not be obvious at one zoom level but becomes clear
when zoomed in.
- **Fix**:
- Try comparing your reference data and classified image at different zoom levels to check for subtle
shifts.
- Perform a visual inspection of both images before accuracy assessment.
By addressing these potential issues, you should be able to better align your classification points with
your reference points and achieve more accurate accuracy assessments.
Sure! Let's go step by step, addressing each potential cause of the problem in detail and how you can fix
or avoid it in ERDAS Imagine.
### 1. **Georeferencing Misalignment**
#### What is it?
Georeferencing is the process of aligning your image to a specific coordinate system or real-world
location. If the georeferencing is inaccurate, the land cover classes in your classified image won’t match
their actual locations.
#### How to Fix It:
**Step-by-Step Guide for Proper Georeferencing in ERDAS Imagine**:
1. **Open your Image in ERDAS Imagine**:
- Launch ERDAS Imagine and open the image you want to georeference.
2. **Add Ground Control Points (GCPs)**:
- Go to **Raster** > **Geometric Correction** > **Polynomial**.
- Click on **GCP Tool** and add Ground Control Points (GCPs) by matching points from your
unreferenced image to a well-referenced base map (which could be a topographic map or satellite
image).
- Place your GCPs on recognizable features (e.g., road intersections, rivers, or buildings) in both
images.
3. **Check the RMSE (Root Mean Square Error)**:
- After adding 4 or more GCPs, the software will calculate the **RMSE**. RMSE measures the
difference between your selected points on the unreferenced image and the reference map. A smaller
RMSE indicates better georeferencing accuracy.
- **Aim for an RMSE below 1 pixel** for optimal accuracy. If the RMSE is too high, adjust the
placement of your GCPs.
4. **Transform and Save the Image**:
- Once the RMSE is acceptable, click on **Compute Transformation** and then save your newly
georeferenced image.
**Things to Check**:
- **Use well-distributed GCPs**: Spread your GCPs across the entire image area, not clustered in one
location, for balanced accuracy.
- **Check on zoom**: Zoom in and visually inspect the alignment of the georeferenced image with the
reference map.
---
### 2. **Projection Issues**
#### What is it?
The projection is the system used to represent the earth’s surface on a flat map. If your image and
accuracy assessment points are in different projections, the points may not align properly.
#### How to Fix It:
**Step-by-Step Guide to Fix Projection Issues in ERDAS Imagine**:
1. **Check the Projection of Your Image**:
- Right-click on your image in the ERDAS Imagine **Layer Manager** and select **Metadata**.
- Under the **Spatial** tab, you will see the projection information of your image. Verify the
**projection type** (e.g., UTM, WGS84) and the **zone** information.
2. **Check the Projection of Your Reference Data (Accuracy Points)**:
- Open your reference data (points or a shapefile) in ERDAS Imagine or any GIS software (like ArcGIS or
QGIS).
- Check the projection information under the **Layer Properties** (in ArcGIS or QGIS) or in the
**Metadata** tab (ERDAS Imagine).
3. **Reproject the Image or Points (if needed)**:
- If the projections do not match, you need to reproject either the image or the reference points.
- Go to **Data Preparation** > **Reprojection** in ERDAS Imagine.
- Choose the source file (image or points) and select the target projection to match the other data.
- Save the reprojected file.
**Important**: Always ensure that all your data (classification results, accuracy points, reference maps)
use the same projection system before proceeding.
---
### 3. **Pixel Shift**
#### What is it?
Pixel shift occurs when slight misalignments between the georeferenced image and reference data
cause points to fall into the wrong land cover classes.
#### How to Fix It:
**Step-by-Step Guide to Minimize Pixel Shift in ERDAS Imagine**:
1. **Check Alignment Before Classification**:
- Before proceeding with classification, zoom in on known features (e.g., roads, rivers) to ensure your
georeferenced image aligns with your reference data.
2. **Use Accurate GCPs**:
- Ensure your GCPs are well distributed and accurately placed, as mentioned in the georeferencing
section. This reduces the likelihood of pixel shifts.
3. **Snap to Grid**:
- In some cases, you can use the **Snap to Grid** function when generating points or performing
classifications. This ensures that the grid of pixels used for classification aligns with the actual locations
on the ground.
4. **Refine Points or Use Buffering**:
- If points still don’t align perfectly, consider creating a small buffer around your accuracy points. This
allows for minor misalignments while still providing valid classifications.
---
### 4. **Image Resampling**
#### What is it?
Resampling occurs when you change the pixel size or geometry of your image, often during
georeferencing or reprojection. Some resampling methods can distort your classification.
#### How to Fix It:
**Step-by-Step Guide for Appropriate Resampling in ERDAS Imagine**:
1. **Choose the Right Resampling Method**:
- When georeferencing or reprojecting, you will be prompted to select a resampling method. The three
most common methods are:
- **Nearest Neighbor**: Best for preserving original pixel values (use this for classification purposes).
- **Bilinear Interpolation**: Averages pixel values, which can blur the image.
- **Cubic Convolution**: Uses weighted averages, which can further distort the image.
- **For classification**, always choose **Nearest Neighbor**, as it preserves the original pixel values.
2. **Reclassify if Necessary**:
- If you suspect resampling has distorted your classification, rerun the classification using the
**Nearest Neighbor** resampling method to maintain accuracy.
---
### 5. **Inaccurate Training Data**
#### What is it?
The training data (used for classification) and reference points (used for accuracy assessment) must be
accurate and correctly placed. If they are not, the classifications will not match your validation points.
#### How to Fix It:
**Step-by-Step Guide for Accurate Training Data in ERDAS Imagine**:
1. **Collect High-Quality Training Data**:
- Use well-distributed and clearly defined land cover types for your training data. Ensure the training
samples represent each land cover class accurately.
- Go to **Drawing Tools** > **Create AOI** to draw polygons representing your training data.
2. **Validate Training Data Placement**:
- After creating training data, overlay your training polygons on the image and zoom in to verify they
are accurately placed over the correct land cover types.
3. **Check Reference Points for Accuracy Assessment**:
- When creating reference points for accuracy assessment, use high-quality GPS or other sources to
ensure these points are correctly placed on the image.
- Avoid points that are near class boundaries, as these may overlap multiple classes, leading to
misclassifications.
---
### 6. **Check Image Alignment at Different Scales**
#### What is it?
Sometimes misalignment issues are only noticeable at certain zoom levels. Checking the alignment at
different scales can help identify small misalignments.
#### How to Fix It:
**Step-by-Step Guide to Check Alignment at Multiple Scales**:
1. **Zoom In on Known Features**:
- Before performing classification or accuracy assessment, zoom in on known, clearly identifiable
features (e.g., buildings, roads) to check for alignment.
2. **Check at Different Zoom Levels**:
- Zoom out and in multiple times to ensure that no distortions appear at different scales.
3. **Overlay Reference Layers**:
- Overlay the georeferenced image with a well-referenced map or layer to visually check for any
misalignments.
By following these detailed steps, you can systematically address and resolve the issues you're facing
with accuracy assessment in ERDAS Imagine. Let me know if you need more clarification on any of these
steps!