Drone Image Super-Resolution With Queue-Aware MEC Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing drone-based surveillance systems face instability in wireless transmission due to the dynamic mobility of drones, leading to inconsistent image and video quality, and conventional static super-resolution frameworks fail to adapt to real-time changes in image and video streams.
Innovation Solution
A deep learning-based super-resolution image processing system that dynamically selects an appropriate super-resolution model based on queue stability and processing performance by utilizing multiple drones and mobile edge computing centers, optimizing image data transmission and processing through adaptive model selection and scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single static super-resolution framework is used, then the system structure is simple, but it cannot adapt to real-time changes in image and video streams from drones
Solution Approach 1:
The patent implements dynamic model selection by maintaining multiple super-resolution models with different complexity levels and selecting the appropriate model based on real-time queue stability conditions. The system transitions from a static single-model framework to a dynamic multi-model framework that adapts to changing transmission conditions, thereby resolving the contradiction between adaptability and system complexity.
Solution Approach 2:
The system changes the parameter of model selection based on queue stability metrics. By monitoring queue stability and dynamically adjusting which super-resolution model is applied (from lightweight to heavy-duty models), the system achieves adaptability to real-time changes while managing computational resources efficiently, thus resolving the contradiction between adaptability and complexity.
2Productivity
If compression is applied to improve transmission speed, then transmission efficiency increases, but image and video quality deteriorates
Solution Approach 1:
The patent introduces super-resolution models as intermediary processing components between the compressed transmission and the final image output. These models act as quality restoration mechanisms that compensate for compression artifacts, allowing the system to maintain high transmission speed while recovering image quality through algorithmic enhancement rather than transmitting uncompressed data.
Solution Approach 2:
The system combines multiple super-resolution models with different characteristics to create a composite processing pipeline. By selecting and combining appropriate models based on queue stability, the system achieves both transmission efficiency and quality restoration, effectively resolving the contradiction between transmission speed and image quality through a composite approach.
3Adaptability or versatility
If multiple super-resolution models are maintained for dynamic selection, then adaptability to queue stability improves, but computational complexity increases
Solution Approach 1:
The patent segments the super-resolution processing capability into multiple distinct models with different computational complexities. By dividing the overall processing task into lighter and heavier models that can be selectively applied based on queue stability conditions, the system achieves adaptability without requiring all models to run simultaneously, thus managing computational complexity while maintaining model selection adaptability.
4Adaptability or versatility
If drone mobility is increased for better surveillance coverage, then monitoring capability improves, but wireless transmission stability deteriorates
Solution Approach 1:
The patent implements a feedback mechanism that monitors queue stability at the MEC server and uses this information to dynamically adjust the super-resolution model selection and transmission parameters. This closed-loop feedback allows the system to compensate for transmission instability caused by drone mobility, maintaining reliable image quality delivery despite varying wireless conditions resulting from enhanced surveillance coverage.
Data Source
AI summary
A deep learning-based super-resolution image processing system may comprise an unmanned aerial vehicle system configured to receive an image captured by an unmanned aerial vehicle and allocate the received image data to a transmission queue for transmission; and a Mobile Edge Computing (MEC) center including one or more transceivers each having a reception queue, and configured to receive the image data from the unmanned aerial vehicle system through the one or more transceivers and generate a super-resolution image corresponding to the image data through a deep learning computation based on a super-resolution model. The unmanned aerial vehicle system may determine which transceiver to transmit the image data based on a data amount difference between the transmission queue and the reception queue. In addition, the MEC center may determine which super-resolution model to apply to the image data such that a time-averaged super-resolution performance defined by Lyapunov optimization is maximized using a quality evaluation value when the super-resolution model has been applied to the image data and a queue stability value of the reception queue.