Table 2.

Results of recent radar based Unmanned Aerial Vehicle (UAV) classification methods

Classification Task (Num. of Classes)	Classification Method	Accuracy (%)	Reference
UAV type vs. birds (11)	Eigenpairs of MDS${}^1$ + non linear SVM${}^2$	82 ${}^{*}$	[16]
UAV type vs. birds (11)	MDS with EMD${}^3$ + SVM	89.54 ${}^{*}$	[24]
UAV type vs. birds (11)	MDS with EMD, entropy from EMD + SVM	92.61 ${}^{*}$	[25]
UAV vs. birds (2)	SVD${}^4$ on MDS + SVM	100	[22]
UAV type (2)	SVD on MDS + SVM	96.2	[22]
UAV vs. birds (2)	2D regularized complex log-Fourier transform + Subspace reliability analysis	96.73	[23]
UAV type + localization (66) ${}^{**}$	PCA${}^5$ on MDS + random forest	91.2	[26]
loaded vs. unloaded UAV (3)	MDS handcrafted features + DAC${}^6$	100	[27]
UAV type (3)	PCA on MDS + SVM	97.6	[29]
UAV type vs. birds (4)	Radar polarimetric features + Nearest Neighbor	99.2	[32]
UAV vs. birds (2)	Range Profile Matrix + CNN${}^7$	95	[36]
UAV type (6)	MDS and CVD${}^8$ images + CNN	99.59	[33]
UAV type vs. birds (3)	SCF${}^9$ reference banks + DBN${}^{10}$	90	[34]
UAV type (2)	Learning on IQ${}^{11}$ signal + MLP${}^{12}$	100	[37]
UAV type (3)	Point cloud features + MLP	99.3	[38]
UAV vs. birds (2)	Motion, velocity and RCS${}^{13}$ features + MLP	99	[39]
UAV type vs. birds (3)	Motion, velocity and signature features + SVM	98	[31]

MDS${}^1$: Micro Doppler Signature, SVM${}^2$: Support Vector Machine, EMD${}^3$: Empirical Mode Decomposition, SVD${}^4$: Singular Value Decomposition, PCA${}^5$: Principal Component Analysis, DAC${}^6$: Discriminant Analysis Classifier, CNN${}^7$: Convolutional Neural Network, CVD${}^8$: Cadence Velocity Diagram, SCF${}^{11}$: Spectral Correlation Function, DBN${}^{10}$: Deep Belief Network, IQ${}^{11}$: In-phase and Quadrature, MLP${}^{12}$: Multi Layer Perceptron, RCS${}^{13}$: Radar Cross Section. ${}^{*}$ These numbers stand for comparable dwell time on the order of $<0.25$ s; ${}^{**}$ Two UAV types, with 35 and 31 locations under test respectively.