Neural Network Pattern Recognition Experiments Toward a Fully Automatic Detection of Anomalies in InSAR Time Series of Surface Deformation

Abstract

We present a neural network-based method to detect anomalies in time-dependent surface deformation fields given a set of geodetic images of displacements collected from multiple viewing geometries. The presented methodology is based on a supervised classification approach using combinations of line of sight multitemporal, multi-geometry interferometric synthetic aperture radar (InSAR) time series of displacements. We demonstrate this method with a set of 170 million time series of surface deformation generated for the entire Italian territory and derived from ERS, ENVISAT, and COSMO-SkyMed Synthetic Aperture Radar satellite constellations. We create a training dataset that has been compared with independently validated data and current state-of-the-art classification techniques. Compared to state-of-the-art algorithms, the presented framework provides increased detection accuracy, precision, recall, and reduced processing times for critical infrastructure and landslide monitoring. This study highlights how the proposed approach can accelerate the anomalous points identification step by up to 147 times compared to analytical and other artificial intelligence methods and can be theoretically extended to other geodetic measurements such as GPS, leveling data, or extensometers. Our results indicate that the proposed approach would make the anomaly identification post-processing times negligible when compared to the InSAR time-series processing.

Landslide Susceptibility Mapping of Landslides with Artificial Neural Networks: Multi-Approach Analysis of Backpropagation Algorithm Applying the Neuralnet Package in Cuenca, Ecuador

“…Differential interferometry (DInSAR) was also applied with COSMO-Skymed (Constellation of Small Satellites for Mediterranean Observation) and Copernicus Sentinel-1 images to identify landslides in areas that were difficult access. It is important to mention that SAR data form a two-dimensional matrix containing phase and amplitude information related to backscattered electromagnetic radiation; therefore, SAR interferometry allows the extraction of the phase component [62]. The wide availability of SAR data, the improvement of processing techniques, and the reduction of satellite revisit times have allowed InSAR to be applied for the analysis of a wide range of ground motion phenomena [63].…”

“…With this, the deformation of the selected pixels was evaluated and the results were geocoded. The use of InSAR has advantages since it allows measuring ground displacements with millimeter precision [61,64], ideal for detecting surface changes [62]. It allows easy monitoring of ground movements, considering the scale of analysis and intensity of deformation [63].…”

Landslide Sensitivity and Response to Precipitation Changes in Wet and Dry Climates

“…With a spatial resolution of 30 m or better this product will be well‐suited for identifying and monitoring landslides. Thus, it is important to continue to explore new approaches to analyze these InSAR products for scientific research, including use of automated or semi‐automated detection and mapping techniques (e.g., Amatya et al., 2021; Milillo et al., 2021).…”

Roadmap on measurement technologies for next generation structural health monitoring systems

“…Machine learning techniques can be used to process and interpret the growing volume of InSAR available data. Examples are the application of neural network approaches to the analysis of displacement measurements on large scale [290] and the automated detection of anomalous points in displacement time series [291]. Overall, the effective use of MT-InSAR measurements to support buildings and infrastructure condition assessment depends on a clear understanding of the effect of assumptions and input parameters on the data processing and of the structural response connected to the monitored deformations.…”