Landslide Detection via Random Forest on Synthetic Remote Sensing Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current landslide detection methods in remote sensing face challenges such as reliance on manual threshold setting, imbalanced sample distribution, and the need for large training datasets, making it difficult to accurately detect landslides in large-scale remote sensing images.
Innovation Solution
A practical method for landslide detection using Landsat 8 images synthesized from the Google Earth Engine, which involves image synthesis, ice and snow detection, removal of non-potential landslide areas, calculation of spectral and textural features, and building a random forest model for improved precision and practicality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If pixel based detection is used, then detection automation is improved, but spatial characteristics and object correlation are lost
Solution Approach 1:
The patent combines pixel-based automated detection with object-oriented spatial analysis by integrating spectral feature extraction at pixel level with region-based object correlation, thereby maintaining automation while recovering spatial characteristics through multi-scale feature fusion
2Reliability
If object oriented detection is used, then spatial characteristics are improved, but segmentation dependency and detection precision are worsened
Solution Approach 1:
The patent applies segmentation by dividing the detection process into multiple stages: initial pixel-level spectral analysis, intermediate object region identification, and final precision detection, where each stage processes data at appropriate scales to minimize segmentation errors
3Measurement precision
If manual threshold setting is used, then detection precision is improved, but time consumption and labor are increased
Solution Approach 1:
The patent implements self-service by enabling the system to automatically learn optimal detection thresholds from training data through machine learning models, eliminating the need for manual threshold setting while maintaining high detection precision through adaptive parameter optimization
4Extent of automation
If machine learning model is used, then automation is improved, but sample distribution balance requirement is worsened
Solution Approach 1:
The patent addresses sample distribution imbalance by transforming the detection parameters through techniques such as resampling strategies, class weight adjustment, and threshold calibration, allowing the machine learning model to adapt to imbalanced landslide and background samples while maintaining automation
5Adaptability or versatility
If deep learning is used, then feature learning capability is improved, but training sample requirement and parameter complexity are increased
Solution Approach 1:
The patent applies partial action by selectively using deep learning components only where they provide significant advantage, while combining them with simpler traditional image processing methods for other tasks, thereby reducing overall parameter complexity while maintaining strong feature learning capability for landslide detection
Data Source
AI summary
This invention discloses a practical method for landslide detection in large space, which comprises the following steps: image synthesis, ice and snow detection, removal of non-potential landslide area, detection of potential landslide area, feature calculation, landslide detection model construction and precision validation; this invention avoids radiometric correction and outlier by detecting landslide from synthetic image. That guarantees practical applicability of the proposal. Firstly, detecting potential landslides can avoid the imbalanced sample distribution issue between background objects and landslides when training the landslide detection model. The landslide is further detected by building a random forest model based on the spectral features and textural features of potential landslide pixels in different neighboring time domains. It fully considers the changes of objects in different time domains, and lays a foundation for efficient landslide extraction. This model is relatively reliable and practical for automatically detecting landslide from large-scale images.