Neuromorphic SNN Image Compression for Low-Power LEO Satellites
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge communication systems face challenges in reducing power consumption and data transmission volume, particularly when handling high-resolution images, due to the inefficiencies in current lossless compression techniques like Arithmetic encoding and LZW algorithms, which require dynamic dictionary management leading to increased processing time and power consumption.
Innovation Solution
The proposed method employs a Spiking Neural Network (SNN) running on neuromorphic edge hardware for filtering and lossless compression of high-resolution images using a bit-packing technique. This approach splits images into grids, identifies boundary pixels, generates data packets, and applies SNN predictions in parallel to reduce data volume and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If traditional lossless compression techniques (Arithmetic encoding, LZW) are used on high-resolution images, then data transmission volume is reduced, but processing time and power consumption increase due to dynamic dictionary management
Solution Approach 1:
The image is divided into multiple patches or blocks, and compression is performed independently on each patch. This segmentation eliminates the need for global dynamic dictionary management while maintaining compression effectiveness, thereby reducing power consumption at the edge device.
Solution Approach 2:
Instead of maintaining and updating complex dynamic dictionaries throughout the compression process, the patent uses simpler, disposable compression models for each image patch. These lightweight models can be discarded after use, avoiding the cumulative power cost of dictionary management.
2Quantity of substance
If traditional lossless compression techniques (Arithmetic encoding, LZW) are used on high-resolution images, then data transmission volume is reduced, but processing time increases due to dynamic dictionary management
Solution Approach 1:
By dividing the image into independent patches and applying compression locally, the patent eliminates the time-consuming global dictionary updates required by traditional methods. Each patch can be compressed independently and in parallel, significantly reducing total processing time.
Solution Approach 2:
The patent pre-trains compression models on representative image data before actual compression. This preliminary action allows the models to be ready for immediate use without requiring complex runtime dictionary construction, thereby reducing processing time during actual compression operations.
3Use of energy by moving object
If SNN-based compression is used at edge devices, then power consumption is reduced, but compression ratio may be limited compared to traditional methods
Solution Approach 1:
The patent applies different compression strategies to different regions of the image based on their importance and characteristics. Critical regions use more aggressive compression while important details maintain higher fidelity, achieving good compression ratios without requiring high power consumption across the entire image.
Solution Approach 2:
The patent applies compression selectively to certain image patches rather than uniformly across the entire image. By focusing compression efforts on less critical regions and using simpler representation for important areas, the system achieves acceptable overall compression ratios while minimizing power consumption at the edge device.
Data Source
Figure 1A
Figure 1B
Figure 2
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 1000x less power compared to similar Artificial neural network running on GPU.