Generate Training Data
Feature Description
When no model is available, the Generate Training Data tool can produce training data for model training using original remote sensing image data and vector annotation data. Due to the large size of remote sensing imagery and the high memory requirements of convolutional neural networks, imagery needs to be processed into machine learning-friendly formats. This tool achieves this by synchronously splitting image data and labels into small tiles, generating neural network-optimized training data.
For "super-resolution reconstruction" training purposes, only low-resolution sample images are required without sample tags.
Parameter Description
| Parameter Name | Default Value | Description | Type |
|---|---|---|---|
| Training Purpose | object detection | Specifies the purpose of training data | Object |
| File Type | Dataset | Specifies input type. For "general change detection", choose Dataset or File List; other purposes support Dataset or Folder | String |
| Sample Image - Datasource | Specifies sample image datasource, supports file database | DataSource | |
| Sample Image - Dataset | Specifies sample image dataset. For "general change detection" only image datasets are supported; other purposes support image/mosaic datasets | Dataset | |
| Sample Image - Folder | Specifies sample image folder path | String | |
| Sample Image - List Path | Specifies image list path (txt/csv format). Each line corresponds to an image path, matching compare image list entries | String | |
| Sample Tag - Datasource | Specifies sample tag datasource | DataSource | |
| Sample Tag - Dataset | Specifies sample label dataset | Dataset | |
| Class Field | Specifies classification field | String | |
| Compare Image - Datasource | Specifies comparison image datasource, supports file database | DataSource | |
| Compare Image - Dataset | Specifies comparison image dataset. For "general change detection" only image datasets are supported; other purposes support image/mosaic datasets | Dataset | |
| Compare Image - List Path | Specifies comparison image list path (txt/csv format). Each line corresponds to a compare image path, matching sample image list entries | String | |
| Rows/Columns | 1024 | Specifies tile size for training data generation | Integer |
| Other Parameter Settings (Optional) |
False | Enables format, offset, and index settings when checked | Boolean |
| Image Format | Training data format: TIFF, JPEG, or PNG | String | |
| Vertical/Horizontal Offset: | 0 | Overlap offset size for expanded training data. Recommended offset is half tile size | Integer |
| Index Start Value | -1 | Default -1 appends new data to existing training data in output directory | Integer |
| Save Unlabeled Data | False | When checked, saves unlabeled tiles for negative sample supplementation | Boolean |
| Export Directory | Specifies output path for generated data | String | |
| Data Name | training_data | Specifies output folder name | String |
Output
Training data folder.