Deep Convolutional Neural Network for Real-Time Crack Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computer vision methods face challenges in real-time crack detection in structures from image data due to high computational burdens and low accuracy, especially with large volumes of video stream data from boroscope cameras and other inspection tools.
Innovation Solution
The use of a deep convolutional neural network (DCNN) for image data processing, which detects regions of interest, applies post-processing to classify defects, and provides a visual indication of defect classification, including a heat map to highlight defect locations and severity, trained using labeled image frames and capable of operating on cloud computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional computer vision algorithms are used for crack detection, then the system can process image data, but the accuracy of defect classification is low and computational burden is high
Solution Approach 1:
The patent segments the image processing task into multiple stages: region of interest detection, patch extraction, and deep neural network classification. This segmentation allows the system to focus computational resources only on relevant areas containing potential defects, reducing overall computational burden while maintaining high classification accuracy through specialized processing at each stage
Solution Approach 2:
The patent introduces a deep convolutional neural network as an intermediary between raw image data and defect classification results. This intermediary layer automatically learns and extracts relevant features from image patches, enabling high accuracy classification without requiring complex manual feature engineering or excessive computational resources for traditional image processing operations
2Productivity
If conventional image processing methods are applied to large volumes of video stream data, then processing can be performed, but real-time detection efficiency is reduced due to high computational overhead
Solution Approach 1:
The patent extracts and processes only regions of interest containing potential defects rather than analyzing entire video frames. By detecting candidate regions first and then applying computationally intensive deep learning only to these extracted patches, the system achieves real-time processing efficiency while maintaining high detection accuracy, significantly reducing computational overhead compared to full-frame processing
Solution Approach 2:
The patent applies partial processing by using lightweight region detection algorithms for initial screening and reserving deep neural network processing only for promising candidate regions. This partial application of computational resources to subsets of data enables real-time processing of large video streams without requiring excessive computational power for every frame
Data Source
AI summary
A method includes detecting at least one region of interest in a frame of image data. One or more patches of interest are detected in the frame of image data based on detecting the at least one region of interest. A model including a deep convolutional neural network is applied to the one or more patches of interest. Post-processing of a result of applying the model is performed to produce a post-processing result for the one or more patches of interest. A visual indication of a classification of defects in a structure is output based on the result of the post-processing.