Recursive Neural Network for Infrastructure Change Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting damage to infrastructure assets, such as electric utility towers and roads, are inefficient and inaccurate, relying on human inspection and machine learning that requires extensive training on every possible component and damage state.
Innovation Solution
The use of image processing techniques, specifically multitemporal image change detection, involving image co-registration, segmentation, and analysis by a recursive neural network, to identify changes in infrastructure assets over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional human inspection methods are used to detect damage to infrastructure assets, then operational simplicity is maintained, but detection accuracy and efficiency deteriorate
Solution Approach 1:
The patent replaces manual human inspection with an automated image processing system that captures images of infrastructure assets and uses recursive neural networks to detect changes and damage. This substitution of mechanical human labor with automated optical and computational systems simultaneously improves both detection accuracy through algorithmic precision and productivity through rapid automated processing of multiple images.
2Adaptability or versatility
If traditional machine learning methods are used for damage detection, then adaptability to various damage states is improved, but training complexity and time requirements worsen
Solution Approach 1:
The patent employs a recursive neural network architecture that is pre-trained on diverse infrastructure images to recognize various damage states. This preliminary training action enables the system to adapt to different damage types without requiring extensive retraining for each specific case, thereby maintaining versatility while reducing the complexity and time required for ongoing adaptation.
Solution Approach 2:
The system changes the parameter of the neural network architecture to use a recursive structure with specific layer configurations (convolutional layers, recurrent layers, fully-connected layers). This parameter change in the model architecture enables efficient processing and reduces training complexity while maintaining the ability to detect various damage states through hierarchical feature learning.
3Measurement precision
If comprehensive image analysis of all infrastructure components is performed, then detection coverage is improved, but processing time and computational resources worsen
Solution Approach 1:
The patent segments the image processing task by dividing infrastructure images into smaller regions or tiles that are processed independently by the recursive neural network. This segmentation allows comprehensive analysis of all components while reducing the computational burden on each processing unit, thereby maintaining high detection accuracy across the entire infrastructure asset while reducing overall processing time through parallel processing of segmented regions.
Solution Approach 2:
The system performs change detection by comparing current images with reference images, focusing computational resources on detecting changes rather than analyzing every pixel of every image from scratch. This partial action approach, where only changes are processed in detail while unchanged regions are efficiently identified through differential processing, reduces processing time while maintaining comprehensive detection coverage.
Data Source
AI summary
Devices, systems and methods that are configured to use image processing to detect structural changes in infrastructure assets are described. An example method for identifying damage in an infrastructure asset includes receiving multitemporal image sets including a time-1 image set and a time-n image set that is sequentially later in time than the time-1 image set, performing an image co-registration operation between the multitemporal image sets and images in a repeat station imaging dataset to generate a registered image pair set, segmenting each image of each image pair of the registered image pair set to generate a plurality of paired tiles, and performing, using a recursive neural network, a change detection operation on each of the plurality of paired tiles.