Spiking Neural Network Lossless Image Compression for LEO Satellites
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low Earth Orbit (LEO) satellites face challenges in processing and transmitting high-resolution earth observation images due to limited memory, computing resources, and power constraints, which hinder real-time analytics and introduce noise artifacts with lossy compression techniques.
Innovation Solution
A method and system utilizing a highly energy-efficient Spiking Neural Network (SNN) for lossless image compression, where neighborhood pixels are normalized and encoded into spike trains, and residual errors are compressed using a classical Arithmetic Encoder, enabling efficient transmission and reconstruction of high-resolution images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If lossy image compression techniques are used to reduce data volume, then downstream data volume is reduced effectively, but noise artifacts are introduced and image details are lost resulting in poor observation results
Solution Approach 1:
The patent transforms image data from continuous pixel values to discrete spike train representations, fundamentally changing the data parameter format. This transformation enables lossless compression by representing image information in a format that preserves all original details while reducing data volume through efficient encoding of the spike train sequences.
2Speed
If high-resolution images are transmitted to receiving stations for further analysis, then real-time analytics capability is maintained, but power consumption for transmission increases and communication delays occur
Solution Approach 1:
The patent segments the image data into local patches and processes each patch independently through the SNN model. This segmentation enables on-board processing of image portions without requiring transmission of the entire high-resolution image, reducing transmission power while maintaining real-time analytics capability for segmented regions of interest.
Solution Approach 2:
The patent performs preliminary compression and processing of image data on the satellite using the SNN model before transmission. By pre-processing and compressing images on-board, the system reduces the data volume that needs to be transmitted, thereby lowering transmission power requirements while preserving essential image information for subsequent analysis.
3Extent of automation
If small satellites are equipped with large computing platforms for on-board processing, then Orbital Edge Computing capability is improved, but power consumption increases and weight restrictions are violated
Solution Approach 1:
The patent replaces traditional heavy mechanical computing hardware with a Spiking Neural Network model that runs on lightweight, low-power neuromorphic or conventional processors. This substitution enables sophisticated on-board image processing and compression without requiring large, power-hungry computing platforms, thus maintaining automation capability while respecting power and weight constraints.
Solution Approach 2:
The patent changes the computational parameter regime by using spike-based temporal coding instead of continuous analog or digital processing. This parameter change enables the SNN to perform complex image processing tasks with significantly lower power consumption and computational resources compared to traditional computing platforms.
Data Source
AI summary
This disclosure relates generally to reducing earth-bound image volume with an efficient lossless compression technique. The embodiment thus provides a method and system for reducing earth-bound image volume based on a Spiking Neural Network (SNN) model. Moreover, the embodiments herein further provide a complete lossless compression framework comprises of a SNN-based Density Estimator (DE) followed by a classical Arithmetic Encoder (AE). The SNN model is used to obtain residual errors which are compressed by AE and thereafter transmitted to the receiving station. While reducing the power consumption during transmission by similar percentages, the system also saves in-situ computation power as it uses SNN based DE compared to its Deep Neural Network (DNN) counterpart. The SNN model has a lower memory footprint compared to a corresponding Arithmetic Neural Network (ANN) model and lower latency, which exactly fit the requirement for on-board computation in small satellite.