Object Extraction

Object extraction utilizes neural network models to preselect regions of interest within imagery, then classifies pixels within these regions to obtain object boundary information, thereby supporting object-level spatial analysis.

Main Parameters

The following figure demonstrates object extraction results of photovoltaic panels in a specified area. Using the instance segmentation model, independent photovoltaic panels are differentiated and extracted to facilitate subsequent statistical analysis.

  1. Function Entry: Toolbox-> Machine Learning-> Imagery Analysis-> Object Extraction tool.
  2. File Type: Supports dataset or folder selection. Enables batch processing to improve inference efficiency. Default: Dataset.
  3. Datasource/Dataset: Required when File Type selects Dataset. Choose the image or mosaic dataset containing target objects.
  4. File Path: Required when File Type selects Folder. Specify the folder path containing *.tif, *.img format images for automatic reading.
  5. Model File: Select object extraction model file (*.sdm).
  6. Probability Threshold: Filters detected objects by confidence score. Only retains objects with scores exceeding this threshold. Default: 0.5.
  7. Deduplicate Threshold: Typically 0.3-0.7. Default: 0.3. Resolves overlapping candidate boxes via Non-Maximum Suppression (NMS), removing boxes with intersection over union (IoU) ratios exceeding this threshold.
  8. Single Step Operation Amount: Number of images processed per inference batch. Default: 1. Higher values increase memory usage but reduce processing time.
  9. Processor Type: Choose between CPU or GPU processing. GPU delivers faster computation.
  10. GPU Number: Specifies GPU device ID for processing. Default: 0. Supports multi-GPU inference.
  11. Other Parameter Settings: Enables additional parameters:
    • Bounds Dataset/Datasource: Performs custom bounds inference using vector dataset extents.
    • Return: When checked, generates minimum bounding box dataset for results (suffix: "bbox").
  12. Result Data: Sets output datasource and dataset name.
  13. Run: Click to execute extraction. Progress information displays in Python window.

Related Content

Machine Learning Environment Configuration

Training Data Generation

Model Training

Object Detection

Binary Classification

Multiple Classification

Scene Classification

Common Change Detection