基于中心注意力机制的无人机个体识别

打开文本图片集

中图分类号：TP3；V 279+.2 文献标志码：B

文章编码：1672-7274（2025）11-0019-04

Individual Recognition of Drones Based on Central Attention Mechanism

YANG Shaofei1²，HU Junjie1，²， JI Yao²， ZHOU Yuxuan1，²，Shen Weigu01，2 （1.National Key Laboratory of Electromagnetic Space Security， Jiaxing 314ooo， China; 2.China Electronics Technology Group Corporation 36th Research Institute，Jiaxing 314ooo，China）

Abstract： The current individual recognition algorithms using drone image transmisson signal time-frequency maps focus onthe entire time-frequency map rather than the signal area in the image，resulting in insufficient recognition accuracy and dificulty in recognizing some drone individuals.Therefore，this paper proposes a drone individualrecognition algorithm based oncentral atention mechanism.This algorithmutilizes thecentral attention mechanism to weighttheregion where the signal is located in the image toenchance the deep neural network model's atention tothatregion.Inorder to improve the algorithm'sabilitytoclassifyindividuals with high similarityand enhance model convergence，alossfunctioncombining focus lossand penalty mechanism is designed.Wecollctand constructindividual recognition datasets fordrones inboth indoorandoutdoorscenarios，and train the modelonthese datasets.The result shows that the proposed algorithm achieves an average recognition accuracy of 98.73% for18 individuals across 4 types of drones，validating the effectiveness of the proposed algorithm.

Keywords：centralatention mechanism;individual recognitionofdrones; focus loss; penalty mechanism;timefrequency map

研究背景

近年来，无人机迅猛发展，在航拍、交通监管及救灾等领域得到了广泛应用[1-4]。（剩余5950字）