Fig 1.
Framework and process flow in the MLDE method.
(a) shows the raw sonar array signal. (b) is the pre-processing of the data after confirming the target, azimuth weighting, and generating the multi-beam LOFAR spectrum, which can be used for CNN model training. (c) is the training of the CNN model. (d) The trained CNN model is used to identify the target and give the azimuth measurements. (e) The azimuth measurements and the trajectory are correlated and filtered by EKF.
Fig 2.
Target multi-beam LOFAR spectrogram obtained in a single collection.
(a) Spectrogram of spatial energy distribution obtained by beamforming, (b) The frequency domain data waveform after target bearing weighting, (c) Time-domain data waveform after target azimuth weighting, (d) Multi-beam LOFAR spectrogram of the target.
Fig 3.
CNN model structure schematic diagram.
Fig 4.
LOFAR spectrum of different samples.
Convert raw sound data in ShipsEar database to LOFAR spectrogram.
Fig 5.
Accuracy and loss of CNN model when training 50 times.
Fig 6.
Accuracy and loss of CNN model when training 100 times.
Fig 7.
Accuracy and loss of CNN model when training 200 times.
Fig 8.
Schematic diagram of the composition of the ship’s radiated noise spectrum.
Fig 9.
Time azimuth histogram obtained from processing simulation data.
Fig 10.
Simulated target azimuth over time.
Fig 11.
Simulation tracking results of the EKF-Max method.
Fig 12.
Simulation tracking results of the MLDE method.
Table 1.
Comparison of tracking azimuth deviation.
Fig 13.
Time azimuth histogram of sea trial data obtained directly using beamforming.
Fig 14.
MFSS at time point A(70s).
Fig 15.
MFSS at time point B(410s).
Fig 16.
MFSS at time point C(600s).
Fig 17.
MFSS at time point D(1020s).
Fig 18.
LOFAR spectrograms from sonar array raw data.
Fig 19.
The spatial azimuth energy distribution of the original data of the sonar array at a certain moment obtained by beamforming.
Fig 20.
Frequency domain beam data for target.
Fig 21.
Frequency domain beam data of the target.
Fig 22.
Multi-beam LOFAR spectrogram of the target.
Fig 23.
Tracking results of the EKF-Max method.
Fig 24.
The tracking results of the MLDE method successfully identify the target where the two azimuth trajectories cross.