Classify Tunnel Points

Function Overview

Classify Tunnel Points is designed for 3D laser point cloud data of mine tunnels. Based on a deep learning model, it performs automatic semantic classification of point clouds and maps original point cloud classes to five target classes: tunnel wall, ventilation duct, cable, debris, and noise. GPU acceleration is supported for batch inference.

Usage

Click Mine > Classify Tunnel Points.

Parameter Settings

• Point Cloud File: Add tunnel point clouds. Multiple files can be queued for processing.

• Source Class: Select the source class involved in classification. The model only predicts points in the selected source classes. Points outside the selected source classes keep their original classes.

• Batch Size: Number of point cloud sampling blocks processed by the model at one time. A larger value loads more data into GPU memory or system memory. The default value is 1.

• Advanced Options: Five models are supported: miou, macc, acc, latest. acc (overall accuracy) reflects overall classification correctness but is sensitive to class imbalance. macc (mean class accuracy) averages accuracy across classes and better evaluates minority classes. miou (mean intersection over union) is a core semantic segmentation metric that evaluates both class prediction and spatial boundary accuracy. A higher value indicates more reliable segmentation. latest is the most recently saved checkpoint and is not guaranteed to have the best performance. miou is recommended.

• Target Class: Five target classes are supported: tunnel wall, ventilation duct, cable, debris, and noise.

• Prefer GPU: Prefer GPU for inference.

Note:

This function overwrites the original data file.