Semi-supervised underwater acoustic source localization based on residual convolutional autoencoder

EURASIP Journal on Advances in Signal Processing

Table 3 Localization performance of broadband dataset

Frequency points	Percentage of label data	75%		37.5%		15%
Frequency points	Positioning network model	MAE (km)	\(P_{CL - 5\% }\)_(%)	MAE (km)	\(P_{CL - 5\% }\)_(%)	MAE (km)	\(P_{CL - 5\% }\)_(%)
①	RA-CAE-SSL	0.3964	82	0.3891	82.5	0.7024	70.5
	CAE-SSL	0.4657	80	0.6604	73.5	0.8721	64.5
	RA-SL	0.4426	81	0.7089	71	0.9729	66.5
	CNN	0.5115	79	0.7859	65	0.6600	65.5
②	RA-CAE-SSL	0.0724	96.5	0.2017	88.5	0.4961	78.5
	CAE-SSL	0.1474	93.5	0.2917	87	0.6400	77
	RA-SL	0.2232	90.5	0.4707	83.5	0.8701	71
	CNN	0.3506	81.5	0.4183	82.5	0.7301	65.5
③	RA-CAE-SSL	0.0100	100	0.0250	98.5	0.1589	91.5
	CAE-SSL	0.0439	98	0.0839	96.5	0.3671	89
	RA-SL	0.0535	98.5	0.0543	97	0.6296	82
	CNN	0.0252	98.5	0.2213	91	.0.5442	82
④	RA-CAE-SSL	0.0077	100	0.0115	99.5	0.0219	99
	CAE-SSL	0.0350	99	0.1135	94	0.1435	92
	RA-SL	0.0177	99.5	0.0466	97	0.1455	94
	CNN	0.0381	98.5	0.0562	96.5	0.2309	88.5

Where①: \({{\{ 109}},{{127\} }}\); ②: \({{\{ 109}},{127},{{145\} }}\); ③: \({{\{ 109}},{127},{{145}},{163},{{198\} }}\); ④: \({{\{ 109}},{127}{{145}},{163},{{198}},{232},{{280\} }}\).