Hierarchical Context Fusion for Multi-Scale Parcel Risk Assessment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing risk assessment systems for natural disasters, such as wildfires, struggle to efficiently account for features at multiple scales using processor- and memory-intensive techniques like feature pyramids, which provide marginal improvements with additional technical costs.
Innovation Solution
A machine learning system utilizing multiple ML models, each associated with a specific image scale and resolution, processes parcel-, neighborhood-, and landscape-level images to generate feature embeddings, which are fused to provide a comprehensive risk assessment.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If feature pyramids are used to account for features at multiple scales, then measurement precision is improved, but device complexity and resource consumption increase significantly
Solution Approach 1:
The patent divides the risk assessment system into multiple independent ML models, each specialized for a specific image scale (parcel-level, neighborhood-level, landscape-level). This segmentation allows each model to focus on features at its designated scale without the complexity of processing all scales, thereby maintaining measurement precision while reducing overall system complexity.
Solution Approach 2:
Each ML model in the patent serves multiple functions: it processes images at its specific scale, generates feature embeddings, and contributes to the overall risk assessment. This multi-functionality eliminates the need for separate processing pipelines for different scales, reducing device complexity while maintaining comprehensive risk assessment accuracy.
2Measurement precision
If feature pyramids are used to account for features at multiple scales, then measurement precision is improved, but processor and memory usage increase
Solution Approach 1:
By segmenting the processing workload across multiple specialized ML models, each handling a specific scale, the patent avoids the redundant processing inherent in feature pyramids. Each model processes only the relevant scale data, reducing overall processor and memory usage while maintaining the ability to assess features at all scales.
Solution Approach 2:
The patent uses separate ML models as functional copies for different scales, where each model is trained specifically for its scale rather than using a single complex model attempting to handle all scales. This approach reduces resource consumption by avoiding the computational overhead of processing all scales through a unified feature pyramid structure.
3Measurement precision
If feature pyramids are used to account for features at multiple scales, then measurement precision is improved, but processing time increases
Solution Approach 1:
The patent segments the multi-scale processing into parallel independent ML models, each handling a specific scale simultaneously. This parallel processing approach eliminates the sequential processing requirements of feature pyramids, reducing processing time while maintaining the ability to capture features at all scales for accurate risk assessment.
Solution Approach 2:
Each ML model is pre-trained on its specific scale data, so during inference, the models can immediately process their respective scales without the computational overhead of building and processing feature pyramids at runtime. This preliminary preparation significantly reduces processing time while maintaining measurement precision.
Data Source
AI summary
Methods, systems, and apparatus for receiving a request for a risk assessment for a parcel, receiving a set of images for the parcel, the set of images including two or more images, each image having an image scale and an image resolution that is different from other images in the set of images, providing a first-level feature embedding and a second-level feature embedding, the first-level feature embedding being provided by processing a first-level image through a first-level machine learning (ML) model, and the second-level feature embedding being provided by processing a second-level image through a second-level ML model, determining a risk assessment at least partially by processing each of the first-level feature embedding and a second-level feature embedding through a fusion network, and providing a representation of the risk assessment for display.