SNN Edge Image Compression for Low-Power Satellite Transmission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge computing systems face challenges in reducing power consumption and data transmission time for high-resolution image processing, particularly in scenarios involving neuromorphic platforms and Spiking Neural Networks (SNNs), as conventional lossless compression techniques like Arithmetic encoding and LZW require dynamic dictionary management, leading to overhead and increased power consumption.
Innovation Solution
A method and system utilizing Spiking Neural Networks (SNNs) on neuromorphic hardware for filtering and bit-packing lossless compression, combined with Distributed Diagonal Parallel Fast Reconstruction (DDPFR) at the central station, to efficiently compress and decompress high-resolution images, reducing data transmission and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional lossless compression techniques (Arithmetic encoding, LZW) are used on edge devices, then lossless compression is achieved, but dynamic dictionary management overhead increases power consumption
Solution Approach 1:
The image is divided into multiple patches, and each patch is processed independently by separate CPU cores. This segmentation eliminates the need for global dynamic dictionary management while maintaining lossless compression, as each core maintains its own local context without requiring synchronization for dictionary updates.
Solution Approach 2:
Each CPU core independently processes its assigned image patch using local context information, making autonomous decisions about compression without requiring centralized dictionary management. This self-service approach eliminates the overhead of dynamic dictionary synchronization while achieving lossless compression through local predictive coding.
2Loss of information
If high-resolution images are transmitted without compression, then image quality is preserved, but data transmission time and bandwidth consumption increase
Solution Approach 1:
The patent transforms the image data from raw pixel values to predictive residuals by changing the reference parameter from absolute pixel values to differences between predicted and actual values. This parameter transformation enables significant data compression while preserving all original information, reducing transmission time without losing image quality.
3Productivity
If parallel processing is used for image compression, then processing speed increases, but system complexity increases
Solution Approach 1:
The image is divided into multiple independent patches that can be processed in parallel by separate CPU cores. This segmentation enables parallel processing to increase throughput while keeping each processing unit simple and independent, avoiding the need for complex inter-core coordination mechanisms.
Solution Approach 2:
Each CPU core processes only its assigned portion of the image independently, performing partial action on the complete image. This approach achieves parallel processing speedup without requiring complex synchronization, as each core completes its task autonomously without needing to coordinate with other cores during processing.
Data Source
AI summary
State of the art techniques for reduced data transmission at edge devices while using lossless compression mostly rely on CNNs that are energy consuming networks. However, for edge devices such as LEO satellites the method and system discloses a unique combination of SNN running on a neuromorphic platform for low-power & low-latency. Relevant data of high-resolution images is filtered, and redundant data is discarded. The filtered data is then compressed using but packing based lossless compression before transmitting to a central station for further processing. At the central station a distributed Diagonal Parallel Fast Reconstruction (DDPFR) is applied to reconstruct the image. The system reduces data sent to the central station and reduces battery consumption of satellite for data transmission. SNN running on neuromorphic processor consumes up to 100 to 1000× less power compared to similar Artificial neural network running on GPU.