US 12,353,977 B1
Unmanned aerial vehicle identification method based on blind source separation and deep learning
Zhigang Zhou, Hangzhou (CN); Jiangong Ni, Hangzhou (CN); Jingyu Zhao, Hangzhou (CN); Xiaona Xue, Hangzhou (CN); Yejiang Lin, Hangzhou (CN); Dou Pei, Hangzhou (CN); and Zhiqun Cheng, Hangzhou (CN)
Assigned to Hangzhou Dianzi University, Hangzhou (CN)
Filed by Hangzhou Dianzi University, Hangzhou (CN)
Filed on Jan. 16, 2025, as Appl. No. 19/023,430.
Int. Cl. G06N 3/04 (2023.01); G06N 3/0442 (2023.01); G06N 3/048 (2023.01); G06N 3/09 (2023.01); H04B 7/185 (2006.01)
CPC G06N 3/0442 (2023.01) [G06N 3/048 (2023.01); G06N 3/09 (2023.01); H04B 7/18504 (2013.01)] 6 Claims
OG exemplary drawing
 
1. An unmanned aerial vehicle (UAV) identification method based on blind source separation and deep learning, comprising the following steps:
S1, acquiring one-dimensional radar cross section millimeter-wave data sets of eight types of UAV, obtaining mixed signals by mixing two sets of four types of data sets with one set being randomly selected of the eight types of UAV and the other set comprising the rest of the eight types of UAV, and using an improved fast independent component analysis (FastICA) algorithm to separate the mixed signals to obtain separated signals;
S2, converting the separated signals into a two-dimensional image by using a data transformation method, augmenting the two-dimensional image to obtain a data set, and dividing the data set into a training set, a validation set, and a test set;
S3, training a UAV classification network model based on a residual network (ResNet18);
S4, carrying out a training of the UAV classification network model in the training set, and obtaining a network model for realizing UAV classification; and
S5, applying the UAV classification network model to classify and identify the UAV.