A Lightweight Collaborative Target Recognition Method for Autonomous Aerial Vehicle Cluster

Abstract

This article aims to explore a reliable target recognition technique for autonomous aerial vehicles (AAV) clusters, and proposes a lightweight collaborative target recognition methods based on multiple viewpoints. In the proposed method, the AAVs are divided into the leaf node AAVs and the head node AAVs. Leaf node AAVs are used for multiview image acquisition and image feature extraction by utilizing a lightweight feature extraction model. The head node AAVs realize efficient feature fusion from the collected image features by using graph convolutional network and graph coarsening techniques. Based on the above lightweight technologies, the proposed method can realize the efficient and accurate target recognition and reduce the demand for limited computing resources and communication resources in AAV clusters. Experimental results show that, compared with existing multiview object recognition method, the proposed method has less computing cost and communication overhead while ensuring reliable recognition accuracy.