Instance matching quality evaluation method and system for point cloud single tree segmentation result

CN122265776APending Publication Date: 2026-06-23SOUTHWEST FORESTRY UNIVERSITY

View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner
Filing Date: 2026-05-15
Publication Date: 2026-06-23

AI Technical Summary

Technical Problem

Existing methods for evaluating single-tree segmentation results are insufficient to fully express the complex spatial overlap between reference and predicted instances in complex forest scenarios, and are also difficult to explain the spatial origins of low-quality matching.

Method used

By acquiring standardized reference point clouds and predicted point clouds, a point-level spatial correspondence is established, a statistical matrix of the intersection of reference instances and predicted instances is constructed, an instance matching quality evaluation index is calculated, and a single-tree segmentation error pattern is identified.

Benefits of technology

It can more fully express the overlapping relationships of instance spaces in complex forest scenarios, explain error patterns, and achieve the integration of quantitative evaluation and error source analysis, making it suitable for unified evaluation of different algorithms.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122265776A_ABST

Patent Text Reader

Abstract

The application provides a point cloud single tree segmentation result-oriented instance matching quality evaluation method and system, belonging to the technical field of forest resource investigation and point cloud processing, and using point-level spatial correspondence to establish overlapping evidence between reference instances and predicted instances. Through reference dominant instance matching relationship, the phenomenon that multiple reference instances correspond to the same predicted instance is reserved. Combined with intersection-over-union grading, predicted instance sharing relationship, reference-to-predicted relationship and predicted-to-reference relationship, not only quantitative evaluation indexes of tree scale and plot scale can be output, but also error modes such as unmatching, weak overlap, instance merging, instance splitting, adjacent tree mixing and prediction competition can be further explained, realizing the integration of quantitative evaluation and error source analysis. The application does not depend on the specific single tree segmentation model structure, as long as the input is a predicted point cloud with instance labels, the unified evaluation of the output results of different algorithms can be realized, and the application has good method independence and complex forest scene adaptability.

Need to check novelty before this filing date? Find Prior Art