AI-Assisted Drawing

Feature Description

AI-assisted drawing is an intelligent object collection tool based on deep learning models. The tool automatically identifies object boundaries based on target features in the imagery, assisting in the collection of areal features such as buildings, roads, water bodies, and vegetation, thereby reducing the workload of manual point-by-point digitization.

It provides two recognition methods: tap recognition and Box selection batch identification. Tap recognition is suitable for extracting single or a few discrete features; box selection batch identification is suitable for batch collection of feature groups in densely built-up areas. After completing AI-assisted extraction, functions such as draw line to modify border and moving line are provided for manual trimming and fine-tuning of the extracted objects.

Application Scenarios

Need to quickly extract boundaries of areal features such as buildings, water bodies, farmland, and vegetation from imagery.
Need to batch collect a large number of areal features in produce data, where traditional manual drawing is inefficient.

Functional Prerequisites

Only supported for drawing on vector polygon layers.
An imagery basemap must be loaded in the map as the source of data for AI recognition.
Need to install the AI extension package and configure the Python environment, otherwise the function is unavailable.

Version Information

This function is supported starting from SuperMap iDesktopX version 2026.

Function Entry

Edit Data tab > Object Drawing group > AI-assisted check box.

Parameter Description

Tap recognition and box selection batch identification share some parameters. The settings dialog box displays corresponding parameters based on the selected recognition method. The parameter description is as follows:

Parameter	Description	Applicable Recognition Method
Recognition Method	Select the recognition method for the target feature. Tap recognition is suitable for single or a few discrete features; box selection batch identification is suitable for batch recognition of multiple features within a rectangular range.	/
Segmentation Threshold	Controls the boundary range of the AI recognition result. The higher the value, the more the recognized boundary shrinks; the lower the value, the wider the recognition range. Value range [0, 1], default is 0.5.	Tap, Box Selection
Smoothness	Controls the smoothness of the recognition result boundary. The larger the value, the more boundary nodes and the smoother the curve. Value range [0, 5], default is 3.	Tap, Box Selection
Recognition Type	Describes the target to be recognized. The model will find and segment the corresponding features within the box selection range. Provides 4 types: Building, Road, Roof, and Custom.	Box Selection Batch Identification Only
Candidate Confidence	Controls the confidence of candidate results for box selection batch identification. Increasing the threshold can reduce low-confidence candidate results; decreasing the threshold can retain more candidate results. Default is 0.17.	Box Selection Batch Identification Only
Mask Threshold	Controls the pixel range of the mask retained by the model. Lowering the threshold usually expands the mask range, potentially reducing missed detections; raising the threshold usually shrinks the mask range, potentially reducing adhesion between adjacent targets. Default is 0.62.	Box Selection Batch Identification Only
Custom Recognition Type	When the Recognition Type is set to "Custom", enter a description of the target to be recognized. Concise and accurate English descriptions are recommended.	Box Selection Batch Identification Only
Automatic Building Regularization	Performs right-angle and regularization processing on the recognized building boundaries, making them closer to rectangles or L-shapes. When unchecked, the original boundary shape from AI recognition is retained. It is recommended to check this when collecting regular buildings.	Tap, Box Selection
Automatic Road Regularization	Performs smoothing and regularization processing on the recognized road boundaries, making them more regular. When unchecked, the original boundary shape from AI recognition is retained. Check this based on the results when collecting roads.	Tap, Box Selection
Frame Selection Fine Recognition	First obtains the center points of each feature in the box selection recognition result, then uses these center points as prompt points for fine recognition to improve object boundaries and recognition completeness.	Box Selection Batch Identification Only
Small-Scale Dense Area Identification	Applicable to box selection recognition of non-road targets. When enabled, the original image is divided into smaller blocks, which helps improve the adhesion and boundary effects of dense buildings; the smaller the block pixels, the longer the recognition time.	Box Selection Batch Identification Only
Block Pixel	Sets the image block size. The smaller the block pixel, the better the recognition edge effect, but the longer the recognition time. Default is 512.	Tap, Box Selection

Operation Steps

Enabling AI-Assisted Drawing

Open the workspace and map, and confirm that the image layer is loaded.
In the layer manager, select the vector polygon layer to which the recognition result will be written, and ensure that this layer is in an editable state.
In the Edit Data tab's Object Drawing group, check the AI-assisted check box. You can also use the shortcut key Alt+Q to quickly toggle the AI-assisted mode.

Tap Recognition

In the Settings dialog box, select tap recognition, and choose appropriate parameters in the Parameter description based on the target feature and recognition effect.
Select the polygon drawing tool: click the Polygon drop-down button, choose "Polygon" or another polygon drawing tool, and move the mouse to the target feature in the imagery.
Click the left mouse button, and AI will automatically recognize and highlight the target feature boundary. If the recognition range is incomplete, continue clicking the left mouse button in the unrecognized area to expand the range.
Hold down the Ctrl key and click the left mouse button to exclude misrecognized areas such as shadows or adjacent features.
Right-click in the map to complete the current object collection. The system automatically performs line smoothing and saves the object to the current editable layer.

Box Selection Batch Identification

Box selection batch identification is suitable for extracting multiple target features within a rectangular range at once.

In the Edit Data tab's Object Drawing group, check the AI-assisted check box, and click the Settings button to open the AI-assisted drawing setup dialog box.
Select Box selection batch identification under Recognition Method.
Choose appropriate parameters in the Parameter description based on the target feature and recognition effect, and click OK to save the parameters.
Select the polygon drawing tool, move the mouse to the target area, hold down the left mouse button and drag to draw a rectangle box covering the target features to be recognized.
Release the left mouse button, and the system will automatically recognize the target features within the box and display the recognition result. The larger the box selection range, the longer the recognition time.
For missed targets, you can switch to tap recognition to supplement. Hold down the Ctrl key and click the left mouse button to deselect objects.
Right-click in the map to complete the current collection and save the recognition result to the current editable layer.

Regularization and Auxiliary Trimming

Regularization is used for post-processing the boundaries of objects after AI recognition. Select one or more AI-assisted polygon objects to be processed, click the regularization drop-down button, and choose Regularize Building or Road regularization based on the object type.
- Regularize Building: Adjusts building boundaries to straight line segments and regularizes right-angle relationships, making the result closer to a rectangle or L-shape.
- Road regularization: Smooths and regularizes road boundaries, reducing minor jitter in the recognition result.
Auxiliary trimming is used to quickly adjust local boundaries based on the AI recognition result. Select the AI-assisted polygon object to be trimmed, click the auxiliary trimming drop-down button, and choose the following tools:
- Draw Line to Modify Border: Draw a line that intersects the boundary of the currently selected polygon object at two or more points, right-click to finish drawing, and the system adjusts the polygon boundary based on the drawn line.
- moving line: Directly drag an edge of the currently selected AI-assisted polygon object to quickly adjust its boundary position.

After trimming is complete, please check the object boundaries and topology, and save the editing results after confirming they are correct.

Note:

The recognition result is affected by image resolution, map scale, target feature texture, and parameters. After recognition, please check the object boundaries and use object editing tools to correct them as needed.
For dense building groups, it is recommended to use box selection batch identification, combined with tap recognition to supplement omissions or exclude misrecognized areas.
When tap recognition and box selection batch identification use the same target layer and object collection type, the recognition result can be written to the current editable layer.
When the recognition mode is box selection batch identification, excluding misselected areas by holding down the Ctrl key and clicking the left mouse button is not supported.