A GRU-Based Method for Predicting Intention of Aerial Targets

Table 6

Intention prediction performance measurement.

Evaluation Index		Precision(%)				Recall (%)				F1 score
		I	II	III	IV	I	II	III	IV	I	II	III	IV
Intent type	Surveillance	83.0	79.2	70.5	68.7	90.0	85.6	77.1	75.4	0.859	0.823	0.737	0.719
	Reconnaissance	89.9	86.8	81.0	76.3	85.3	78.8	75.4	72.4	0.876	0.826	0.781	0.743
	Feint	90.6	89.1	82.5	78.1	85.4	82.9	73.8	75.5	0.879	0.859	0.779	0.768
	Attack	82.3	79.6	71.2	70.2	90.6	89.7	81.4	73.3	0.862	0.843	0.759	0.717
	Penetration	90.3	90.0	83.4	80.8	89.9	89.6	84.9	83.1	0.901	0.897	0.841	0.819
	Retreat	97.5	98.3	93.4	95.3	94.7	93.9	91.5	91.1	0.961	0.960	0.924	0.931
	Electronic jamming	99.5	96.4	96.6	94.4	95.5	96.0	89.7	90.6	0.975	0.962	0.931	0.925

I, II, III, and IV, respectively, represent the BiGRU-Attention, LSTM, SAE, and DBP aerial target air tactical intention recognition models.