Individual Tree Crown Segmentation

Function Overview

This function utilizes deep learning to extract tree crown contours from images for individual tree segmentation. The input must be a multi-channel color image such as DOM (Digital Orthophoto Map) or orthoimages generated from aerial imagery. It cannot be single-channel images like DEM or DSM. The output is a vector file containing multiple crown contours and corresponding individual tree attribute file.The position of trees in the individual tree attribute file is calculated using the average value of contour points.

Usage

Click ALS Forest > Segmentation > Individual Tree Crown Segmentation

Parameter Settings

• Select file list：Choose image data already opened in the software or from a folder. Currently supports 8-bit, at least three-band high-resolution (1-10 cm) tif format images.
• Mask File: Set the region of interest for Individual Tree Crown Segmentation. If this file is added, only trees within the region of interest will be segmented.If the TIFF file has an MLC file, you can select one or more categories. Only the selected categories will be extracted.
• Overlap Ratio（Default: "0.7"）：If the intersection over union (IoU) of two contours exceeds this threshold, the contour with higher confidence is retained.
• Confidence Threshold（Default: "0.3"）：Only retain contours with tree crown confidence higher than this threshold.
• Tree Species：Select the appropriate model based on tree species type; currently includes a general model and a palm tree model.
• Output Path ： Save path for the output contour vector file and individual tree attribute files.

Model Architecture

The model is built based on the YOLO v8 model architecture from the Ultralytics package. The model achieves an mAP (mean Average Precision) of 95%.