A Distributed Convolutional Neural Network Hierarchical Matching Method Based on Unmanned Aerial Vehicle Swarm

Abstract

The invention discloses a distributed convolutional neural network hierarchical matching method based on UAV clusters, which is applied to the field of UAV networks oriented to the Internet of Things. It takes a long time to use the convolutional neural network to identify and calculate the perceived high-definition images. The present invention decouples the convolutional neural network into different layers, and assigns different convolutional neural network layers to each UAV for collaboration. Calculation, effectively reducing the processing time of a single picture on a single UAV, and at the same time comprehensively considering the position and computing power of the UAV through the deep reinforcement learning algorithm, to determine the optimal distributed convolutional neural network for each picture and the UAV. The matching scheme enables multiple pictures to be staggered in the drone cluster for calculation and processing, which effectively improves the utilization of computing resources of the drone and achieves the goal of minimizing the task completion time.

Description

technical field [0001] The invention belongs to the field of unmanned aerial vehicle networks oriented to the Internet of Things, and in particular relates to an image recognition technology for an unmanned aerial vehicle cluster using a convolutional neural network collaborative calculation. Background technique [0002] As a new technology, UAV swarms are widely used in civil fields, including forest fire monitoring, terrain observation, post-disaster rescue, etc. A large number of these applications require UAVs to identify and process the image information captured by the camera. Condition. In order to ensure a wide perception field of view, a perception UAV carries multiple cameras to complete the shooting of suspicious task points from different angles at the same time, and the multiple perception pictures are handed over to the computing UAV for processing. [0003] At present, a more accurate image recognition method is to use convolutional neural network technology...

AI Technical Summary

Problems solved by technology

At this stage, many technologies in the UAV network cannot properly solve the problem. Due to the lack of computing power of UAVs, it takes a long time for a single computing UAV to process multiple images. The convolutional neural network calculation process takes a long time. How to use UAV clusters To collaboratively complete the convolutional neural network calculation process of multiple pictures taken by a perception drone at one time to minimize the task completion time

Images