Transformer-Based Pavement Anomaly Detection Using Patch Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pavement distress detection methods require extensive data processing and storage due to high data volume and repeatability, making them time-consuming and inefficient, especially for low-cost and high-frequency detection.
Innovation Solution
A pavement nondestructive detection and identification method based on small samples using a Transformer model for feature extraction and target reconstruction, which divides images into patch blocks, reduces pixel resolution, and employs a masked autoencoder for pixel-level reconstruction and classification, allowing for efficient data processing and anomaly recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large number of images are collected for pavement distress detection, then detection accuracy is improved, but data storage and processing time increase significantly
Solution Approach 1:
The patent applies preliminary action by performing data filtering and anomaly detection at the acquisition end before data is stored or transmitted. The system pre-processes images to identify and retain only those containing pavement distress, eliminating the need to store and process all collected images. This advance filtering significantly reduces subsequent processing time while maintaining detection accuracy.
Solution Approach 2:
The patent extracts only the essential information from image data by using anomaly detection algorithms to identify and extract images containing pavement distress. Instead of processing complete image datasets, the system extracts and retains only the relevant distress images, reducing data volume for storage and processing while preserving detection accuracy.
2Productivity
If low-cost cameras and high-frequency detection are used, then detection coverage is improved, but data volume and processing pressure increase
Solution Approach 1:
The patent implements preliminary filtering at the data acquisition stage using lightweight anomaly detection algorithms. This pre-processing step identifies and flags only those images containing pavement distress, enabling the system to handle high-frequency detection data from low-cost cameras while minimizing the volume of data requiring storage and detailed processing.
Solution Approach 2:
The patent segments the detection process into multiple stages: initial anomaly detection at acquisition, selective retention of distress images, and focused processing of identified anomalies. This segmentation allows the system to maintain high detection coverage from frequent low-cost imaging while managing data volume by processing only the segmented subset of relevant images.
3Measurement precision
If complete image datasets are processed for pavement distress identification, then identification accuracy is improved, but work efficiency decreases
Solution Approach 1:
The patent applies preliminary anomaly detection to filter the complete image dataset before detailed processing. By pre-identifying images containing pavement distress, the system maintains the ability to use complete image data for accurate identification while improving work efficiency by reducing the volume of data requiring comprehensive processing.
Solution Approach 2:
The patent extracts only the distress-containing images from complete datasets for detailed processing. This extraction approach preserves identification accuracy by maintaining access to complete original images when needed, while improving productivity by focusing computational resources only on the extracted subset of relevant images.
Data Source
AI summary
Provided is a pavement nondestructive detection and identification method based on small samples, including: constructing an original dataset, dividing the original dataset into several patch blocks, sampling the patch blocks, and obtaining samples of the patch blocks; inputting the samples of the patch blocks into a Transformer model for feature extraction and target reconstruction, and obtaining a trained Transformer model; and based on the trained Transformer model, detecting input pavement sample images.