Explainable AI Damage Detection with Monochrome Image Processing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image analysis systems, particularly those using Convolutional Neural Networks (CNNs), are opaque and difficult to explain, leading to low precision and high error rates in damage detection and estimation tasks, and are hindered by data privacy and regulatory issues.

Innovation Solution

An AI-based damage detection and estimation system using an ensemble cause prediction model with multiple sub-models and explainable AI (XAI) techniques to provide visual explanations for damage cause identification, combined with monochrome image processing to improve accuracy and reduce resource usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Extent of automation

If traditional CNNs are used for damage detection, then automation is improved, but measurement precision deteriorates due to opacity and difficulty in explanation

Engineering Contradiction:
ImproveautomationVSAvoidmeasurement precision
Core Design Contradiction:
Extent of automationVSMeasurement precision

Solution Approach 1:

The patent introduces XAI techniques as an intermediary layer between the CNN and the final damage assessment. This intermediary provides visual explanations and interpretable outputs that bridge the gap between automated processing and precise, explainable results, allowing the system to maintain automation while improving measurement precision through transparent decision-making processes.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the damage detection process into multiple interpretable components including cause prediction, part identification, and damage assessment. By breaking down the automated CNN process into discrete, explainable stages with visual outputs at each step, the system maintains automation while improving precision through systematic verification at each segment.

Inventive Principle:
Principle #1Segmentation

2Loss of information

If color images are processed, then information completeness is improved, but use of energy deteriorates due to higher computational requirements

Engineering Contradiction:
Improveinformation completenessVSAvoiduse of energy
Core Design Contradiction:
Loss of informationVSUse of energy by moving object

Solution Approach 1:

The patent extracts and processes only the essential monochrome information from images rather than processing full color data. By taking out the critical structural and damage-related information while discarding non-essential color data, the system reduces energy consumption while maintaining sufficient information completeness for accurate damage detection.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the parameter representation from full-color RGB data to monochrome intensity values. This parameter transformation reduces the data volume and computational complexity by approximately 2-3 times, significantly lowering energy consumption while preserving the essential information needed for damage detection and assessment.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3839822B1Explainable artificial intelligence (AI) based image analytic, automatic damage detection and estimation system
Publication Date: 2025.09.03 ACCENTURE GLOBAL SOLUTIONS LTD
  • EP3839822B1 patent drawingFigure 1
  • EP3839822B1 patent drawingFigure 2
  • EP3839822B1 patent drawingFigure 3

AI summary

An Artificial Intelligence (Al) based automatic damage detection and estimation system receives images of a damaged object. The images are converted into monochrome versions if needed and analyzed by an ensemble machine learning (ML) cause prediction model that includes a plurality of sub-models that are each trained to identify a cause of damage to a corresponding portion for the damaged object from a plurality of causes. In addition, an explanation for the selection of the cause from the plurality of causes is also provided. The explanation includes image portions and pixels of images that enabled the cause prediction model to select the cause of damage. An ML parts identification model is also employed to identify and labels parts of the damaged object which are repairable and parts that are damaged and need replacement. The cost estimation for the repair and restoration of the damaged object can also be generated.