Hierarchical SNN Architecture for Satellite Image Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems for processing high-resolution images on satellites face challenges with power consumption, communication bandwidth, and efficiency, particularly in Bent-Pipe architectures where all image data is transferred to the ground station for filtering.
Innovation Solution
The implementation of an energy-efficient hierarchical multi-stage Spiking Neural Network (SNN) architecture for classification and segmentation of high-resolution images. This approach divides images into smaller patches, classifies them using a neuromorphic platform, and aggregates labels to generate segmentation maps, reducing power consumption and communication bandwidth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If all high-resolution image data is transferred to ground station for filtering (Bent-Pipe architecture), then data processing capability is improved, but communication bandwidth consumption increases
Solution Approach 1:
The patent segments high-resolution images into multiple smaller patches that are processed independently through hierarchical SNN classifiers. This segmentation allows selective processing and transmission of only necessary data, reducing overall communication bandwidth consumption while maintaining processing capability.
Solution Approach 2:
The patent performs preliminary classification and segmentation of image patches on-board the satellite before data transmission to the ground station. By pre-processing and filtering data at the source, the system reduces the volume of data requiring transmission, thereby lowering communication bandwidth requirements.
2Measurement precision
If all high-resolution image data is transferred to ground station for filtering (Bent-Pipe architecture), then data processing capability is improved, but power consumption increases
Solution Approach 1:
By dividing images into smaller patches and processing them through hierarchical SNN classifiers on-board, the system performs computation locally rather than requiring all data to be transmitted and processed at the ground station. This segmentation approach reduces overall power consumption by enabling distributed processing.
Solution Approach 2:
The patent implements preliminary classification and filtering of image data on-board the satellite using SNN processors before transmission to the ground station. This pre-processing reduces the computational burden at the ground station and lowers total system power consumption by performing necessary processing at the edge.
3Device complexity
If traditional Bent-Pipe architecture is used, then system simplicity is maintained, but processing time increases
Solution Approach 1:
The patent performs preliminary classification and segmentation of image patches on-board the satellite before data transmission to the ground station. This pre-processing reduces the volume of data requiring transmission and processing at the ground station, thereby reducing overall processing time while adding only moderate system complexity through on-board SNN processors.
Data Source
AI summary
State of art techniques rely of FPGA based approaches when power efficiency is of concern. However, compared to SNN on Neuromorphic hardware, ANN on FPGA requires higher power and longer design cycles to deploy neural network on hardware accelerators. Embodiments of the present disclosure provide a method and system for energy efficient hierarchical multi-stage SNN architecture for classification and segmentation of high-resolution images. Patch-to-patch-class classification approach is used, where the image is divided into smaller patches, and classified at first stage into multiple labels based on percentage coverage of a parameter of interest, for example, cloud coverage in satellite images. The image portion corresponding to the partially covered patches is divided into further smaller size patches, classified by a binary classifier at second level of classification. Labels across multiple SNN classifier levels are aggregated to identify segmentation map of the input image in accordance with the coverage parameter of interest.