Wrapped Interferogram Analysis for Creeping Landslide Identification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional landslide identification methods using InSAR technology face limitations in large-area rapid response scenarios due to the lack of effective automated analysis techniques, misclassification from residual atmospheric phases, and high computational intensity, leading to suboptimal data utilization and reliance on expert knowledge.
Innovation Solution
An intelligent identification method based on wrapped interferograms, utilizing data preprocessing, multi-branch feature extraction, and a model generalization mechanism to accurately extract creeping landslide areas, involving data acquisition, feature fusion, and an intersection-union decision process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If InSAR technology is used for landslide identification, then detection sensitivity and spatial coverage are improved, but automated analysis capability and real-time response are deteriorated due to lack of effective techniques and over-reliance on expert knowledge
Solution Approach 1:
The system enables automated self-service processing of InSAR data through the neural network model that automatically performs feature extraction, phase unwrapping, and landslide identification without requiring expert intervention. The model processes wrapped interferograms directly through encoding layers, feature fusion mechanisms, and decoding layers to generate identification results autonomously.
Solution Approach 2:
The patent replaces manual expert analysis with an automated neural network system. The mechanical process of expert knowledge application is substituted by an intelligent algorithm that learns from training data and automatically performs landslide identification, transforming the workflow from human-dependent to system-dependent processing.
2Measurement precision
If image denoising techniques are used to segregate atmospheric delay phase in unwrapped interferograms, then displacement recovery is improved, but misclassification occurs due to residual phases exhibiting spatial similarity with deformed phases
Solution Approach 1:
The neural network model extracts and separates atmospheric delay phases from deformation phases through learned feature representations. The encoding layers identify distinctive patterns of atmospheric artifacts and isolate them from true deformation signals, enabling accurate segmentation without the misclassification problems of traditional denoising methods.
Solution Approach 2:
The patent transforms the phase data representation by mapping wrapped interferograms through sine and cosine functions into a different parameter space. This parameter transformation enhances the distinguishability between atmospheric phases and deformation phases, allowing the neural network to more effectively separate these components and improve classification reliability.
3Manufacturing precision
If time-series deformation products are used as network inputs for semantic segmentation, then boundary delineation accuracy is improved, but computational intensity and data preparation time are significantly increased
Solution Approach 1:
Instead of processing complete time-series deformation products, the patent applies partial action by using a single wrapped interferogram as input. The neural network model achieves effective landslide identification and boundary delineation from this partial input, avoiding the excessive computational burden of processing entire time-series data while maintaining acceptable accuracy through the model's feature extraction and fusion capabilities.
4Productivity
If wrapped interferograms are processed directly without phase unwrapping, then computational cost is reduced, but phase information completeness is deteriorated
Solution Approach 1:
The patent applies parameter changes by transforming the wrapped phase values through sine and cosine functions. This transformation preserves the essential phase information in a different representation that the neural network can effectively process, maintaining information completeness while avoiding the computational complexity of phase unwrapping operations.
Solution Approach 2:
The sine and cosine transformations serve as intermediary representations between the raw wrapped interferogram and the neural network processing. These intermediary forms preserve the phase information needed for accurate landslide identification while being computationally more efficient to handle than unwrapped phases, acting as a bridge that maintains information integrity without requiring full phase unwrapping.
Data Source
AI summary
In an intelligent identification method for creeping landslide hazards based on wrapped interferograms is proposed. Multi-scene time-continuous wrapped interferogram data are first acquired and upsampled to 128×128 using bilinear interpolation. Sine and cosine functions are applied to generate sine and cosine phases, which are mapped into the RGB color space for multidimensional representation. A multi-branch network with shared weights extracts features independently, and same-layer features are fused to enhance high-level semantics. Then, the decoder progressively restores the feature information and completes the classification mapping, yielding the identification result of a single-scene interferogram. Finally, the model generalization mechanism based on an intersection and merging decision is applied to produce the final recognition result.