Property Feature Segmentation Using Parcel-Guided Instance Masks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to accurately segment property features in aerial imagery due to under-segmentation issues and the inability of manual labelers to reliably distinguish instances, especially in dense developed environments.
Innovation Solution
A method for property feature instance-aware segmentation that uses parcel data to distinguish between contiguous property features, incorporates artificially inserted air gaps during training, and employs a trained segmentation module to generate pixel-accurate masks, leveraging a combination of semantic and instance-aware segmentation techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is used to distinguish property feature instances, then labeling precision is improved, but labeling time and labor cost increase significantly
Solution Approach 1:
The patent uses automatically generated pseudo-labels as copies of manual annotations. These pseudo-labels are created by processing satellite imagery through a segmentation model that generates property feature instance masks, which then serve as training data without requiring manual verification for each annotation.
Solution Approach 2:
The system performs self-service by using its own outputs (segmentation results) to generate training data. The segmentation model processes imagery, creates property feature instance masks, and uses these masks as training labels, eliminating the need for human annotators to manually create training data.
2Manufacturing precision
If traditional segmentation methods are used, then processing speed is maintained, but segmentation precision deteriorates due to under-segmentation
Solution Approach 1:
The patent applies segmentation by dividing property features into distinct instances based on visual characteristics and spatial relationships. The segmentation model identifies and separates individual property features (buildings, roads, vegetation) within satellite imagery, creating precise instance-level masks that traditional methods fail to distinguish.
Solution Approach 2:
The system applies local quality by analyzing specific local regions of the imagery to determine property feature boundaries. The segmentation model examines local visual patterns, textures, and spatial relationships to accurately delineate individual property features, achieving high precision without requiring processing of entire images at once.
3Measurement precision
If parcel data is integrated into segmentation, then segmentation accuracy is improved, but data processing complexity increases
Solution Approach 1:
The patent uses parcel data as an intermediary that bridges satellite imagery and property feature segmentation. Parcel boundaries and attributes serve as intermediate information that guides the segmentation model in distinguishing property features, particularly in dense developed areas where visual distinction alone is insufficient.
Solution Approach 2:
The segmentation system achieves multi-functionality by integrating multiple data sources (satellite imagery, parcel data, and generated pseudo-labels) into a unified processing framework. The same segmentation model processes both visual information and parcel data to produce consistent property feature instances, making the system adaptable to various data types and purposes.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The method for determining property feature segmentation includes: receiving a region image for a region; determining parcel data for the region; determining a final segmentation output based on the region image and parcel data using a trained segmentation module; optionally generating training data; and training a segmentation module using the training data S500.