UAV Image Offloading for Deep Learning Detection and Noise Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Unmanned air vehicles (UAVs) face limitations due to limited battery power and computational resources, making it challenging to run computationally heavy image processing tasks like object tracking, detection, or segmentation directly on board.
Innovation Solution
A system where images are processed by a control unit on the UAV, modulated, and transmitted to a remote server for further processing using pre-trained deep learning models, reducing computational and power requirements on the UAV by performing noise removal and image processing tasks on the server.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning-based tracking, detection, or segmentation algorithms are used on UAVs, then object monitoring and detection accuracy is improved, but energy consumption and computational requirements increase significantly
Solution Approach 1:
The patent extracts the computationally intensive deep learning processing tasks from the UAV platform and relocates them to a remote server. The UAV only performs lightweight preprocessing and transmits images, while the server handles the heavy computational load of object detection, tracking, and segmentation, thereby resolving the contradiction between detection accuracy and energy consumption.
Solution Approach 2:
The patent introduces a remote server as an intermediary between the UAV and the final processing output. The server acts as a mediator that receives preprocessed images from the UAV, performs the computationally heavy deep learning operations, and returns results, allowing the UAV to maintain high detection accuracy without bearing the energy burden.
2Productivity
If deep learning models are trained and executed on the UAV board, then real-time processing capability is improved, but device complexity and processing requirements increase
Solution Approach 1:
The patent extracts the complex deep learning model execution from the UAV board and moves it to a remote server. The UAV board only handles simple image capture and preprocessing, while the server hosts the full deep learning models for object detection and segmentation, thereby reducing device complexity while maintaining processing capability.
Solution Approach 2:
The patent shifts the processing dimension from local (on-UAV) to remote (server-based). By moving the computational workload to a different spatial dimension (remote server infrastructure), the system achieves real-time processing capabilities without increasing the complexity of the UAV's onboard hardware.
3Device complexity
If images are transmitted to a remote server for processing, then computational requirements on the UAV are reduced, but communication bandwidth and transmission time are required
Solution Approach 1:
The patent applies preliminary action by having the UAV perform image preprocessing (such as compression, filtering, or feature extraction) before transmission. This preparation work reduces the amount of data that needs to be transmitted to the server, thereby minimizing communication bandwidth requirements and transmission delays while still achieving the goal of reducing onboard processing complexity.
Data Source
AI summary
A system includes an air vehicle having an image sensor to acquire images, a control unit to process images acquired by the image sensor, and a communication unit to transmit images processed by the control unit to a remote server by at least one base station. Accordingly, the control unit is configured to modulate the acquired images and send them to the base station by the communication unit; the base station is configured to demodulate the images coming from the air vehicle and to transmit the received images to the server; the server includes a memory unit containing a pre-trained deep learning model to process the images and a processor unit configured to process the images coming from the air vehicle through the deep learning model in the memory unit; the processor unit is configured to perform a noise removal process to reduce the noise.