A joint underwater acoustic target passive detection and bearing estimation method based on YOLO model

By directly outputting the probability of the target's spectrum and its orientation using the YOLO model, the efficiency and accuracy issues of underwater acoustic target detection and orientation estimation in multi-interference environments are solved, achieving efficient target detection and DOA estimation.

CN117214818BActive Publication Date: 2026-07-17INST OF ACOUSTICS CHINESE ACAD OF SCI
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
INST OF ACOUSTICS CHINESE ACAD OF SCI
Filing Date
2022-06-01
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing passive underwater acoustic target detection methods are slow to process and have low information utilization in environments with multiple interferences, resulting in poor target detection and azimuth estimation performance.

Method used

A convolutional neural network based on the YOLO model is used to extract frequency-beam domain features through broadband beamforming. A loss function is designed and the YOLO model is trained to directly output the probability of the target spectrum and its azimuth information, thus avoiding two-stage processing.

Benefits of technology

It improves target detection rate and DOA estimation accuracy, increases processing speed by more than 4 times, achieves a target detection rate close to 100%, reduces DOA mean square error to 0.54°, and outputs results independently to avoid outliers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117214818B_ABST
    Figure CN117214818B_ABST
Patent Text Reader

Abstract

The present application relates to a kind of joint underwater acoustic target passive detection and azimuth estimation method based on YOLO model, the method is first based on the underwater acoustic data received by towed horizontal array, using wideband beam forming method extracts frequency-beam domain FBD feature, and is divided into training set and test set, and sample expansion is carried out to training set;Then, the convolutional neural network of YOLO model is built, loss function is designed, and the FBD feature sample in the training set after expansion is used to train YOLO model;Finally, the FBD feature sample in test set is input into trained YOLO model, to obtain the existence probability of target spectrum in FBD feature and the specific azimuth of target spectrum.The present application can directly input FBD feature into YOLO model, and simultaneously use regression method to obtain target existence probability and target DOA information in FBD sample, effectively utilize the joint information of target spectrum and target azimuth, can detect the existence probability of target source in the marine environment of multiple interference, while giving the specific azimuth of target source.
Need to check novelty before this filing date? Find Prior Art