Building Collapse Material Segmentation With Domain Adaptation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current demolition machinery operates via manual remote control, limiting automation and intelligence in scenarios with diverse materials, necessitating frequent drill bit changes and parameter adjustments, especially in building demolition, where scenario adaptability is weak.

Innovation Solution

A material segmentation method and system employing domain adaptive learning to construct samples from building collapse scenarios, using labeled images and data enhancement techniques to improve model efficiency and accuracy, incorporating domain migration training and fine-tuning to adapt models for specific scenarios.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If manual remote control is used for demolition machinery, then human judgment can be applied, but automation level is limited and operation efficiency is reduced

Engineering Contradiction:
Improveautomation levelVSAvoidoperation efficiency
Core Design Contradiction:
Extent of automationVSProductivity

Solution Approach 1:

The patent replaces manual remote control with an automated material segmentation system using deep learning models and computer vision technology. The system automatically identifies and segments different materials in real-time, substituting human judgment with automated image processing and neural network analysis, thereby achieving both high automation and maintained operation efficiency

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The demolition machinery is equipped with onboard cameras and computing systems that enable it to autonomously perform material identification and segmentation without external human intervention. The system serves itself by automatically adapting working parameters based on real-time material detection, eliminating the need for continuous manual supervision while maintaining high operational efficiency

Inventive Principle:
Principle #25Self-service

2Measurement precision

If deep learning models are trained with manually labeled building collapse images, then material segmentation accuracy can be improved, but sample labeling difficulty and time consumption increase significantly

Engineering Contradiction:
Improvematerial segmentation accuracyVSAvoidsample labeling time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent pre-trains deep learning models using large-scale road scenario material datasets before deploying them to building collapse scenarios. This preliminary training on abundant road material data establishes a strong foundation model that can be quickly adapted to building collapse scenarios with minimal additional labeled data, significantly reducing the time required for scenario-specific labeling while maintaining high segmentation accuracy

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system employs domain adaptive learning to dynamically adjust model parameters when transitioning from road scenarios to building collapse scenarios. By changing the adaptation parameters rather than retraining the entire model from scratch, the system achieves high accuracy in building collapse material segmentation while minimizing the time investment required for scenario-specific data labeling

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If general deep learning models are used for material segmentation, then model construction can be simplified, but scenario adaptability is insufficient for diverse building demolition environments

Engineering Contradiction:
Improvescenario adaptabilityVSAvoidmodel construction complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent segments the model construction process into distinct phases: pre-training on road scenarios, domain adaptation to building scenarios, and fine-tuning for specific collapse types. This segmentation allows each phase to be optimized independently, improving scenario adaptability while managing overall model construction complexity through structured, modular development

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system uses a universal deep learning backbone model that can function across multiple scenarios (road materials, building materials, various collapse types) by applying domain adaptive learning. This multi-functional approach allows a single model architecture to adapt to diverse building demolition environments without requiring separate specialized models for each scenario, thereby enhancing versatility while controlling construction complexity

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP4672169B1Material segmentation method and system for building collapse scene
Publication Date: 2026.05.13 JIANGSU XCMG CONSTRUCTION MACHINERY RESEARCH INSTITUTE LTD
  • EP4672169B1 patent drawingFigure 1~2
  • EP4672169B1 patent drawingFigure 3~4
  • EP4672169B1 patent drawingFigure 5

AI summary

The invention discloses a material segmentation method and system for building collapse scenarios. The method comprises: inputting a target building collapse scenario image to be segmented into a trained building collapse scenario material segmentation model to obtain a material segmentation result; wherein the acquisition of the building collapse scenario material segmentation model comprises: acquiring a road scenario material data set; mapping road scenario material segmentation images in the road scenario material data set to obtain a building scenario material data set; adding collapsed area material objects and corresponding material labels to an image of the building scenario material data set to obtain a building collapse scenario material data set; training a second deep semantic segmentation model using the building collapse scenario material data set to obtain a basic building collapse scenario material segmentation model; and performing domain adaptive learning on the basic building collapse scenario material segmentation model using labeled images in target scenario sequence images to obtain the trained building collapse scenario material segmentation model.