Sonar Target Detection Based On Deep Learning

In recent years,the global international political situation has become more complicated and severe,and the ecological environment in the sea has deteriorated.How to properly develop marine resources and effectively protect the sovereignty of the state territorial waters is an important task facing all countries.In addition,the underwater sonar target detection and identification has been plagued by the domestic and international water sound circles.How to improve the accuracy and speed of underwater sonar target detection is an important research topic in all countries.Under such urgent needs and research background,this paper develops key technologies such as detection and recognition of sonar target based on deep learning.It mainly includes:three-dimensional sonar signal data pre-processing,sonar target feature extraction under low SNR,underwater sonar target detection,location and classification.The feature extraction,detection location and classification of sonar target are mainly studied from the following two aspects:(1)This paper proposes a method for detecting sonar target based on Faster R-CNN.In order to improve the accuracy and speed of underwater target feature extraction,a large amount of underwater acoustic data is used for feature extraction research.This paper attempts to apply the algorithm model of Faster R-CNN to the detection of sonar target,and make relevant adaptation improvements,so that it can better play the advantage of neural network in the detection and recognition of sonar target.Through a large number of sonar data training and test research,the deep learning technology used in this paper can efficiently realize the feature extraction of target in different complex underwater environments.Through the Faster R-CNN,the speed and accuracy of feature extraction of underwater acoustic data are improved under the background of low SNR,which provides a new research method for feature extraction of complex underwater acoustic environment data.(2)This paper proposes a method for detecting sonar target based on SSD.Through the experimental research on the sonar target detection algorithm based on Faster R-CNN,it is found that the technique of deep learning breaks through the bottleneck of the detection and recognition of sonar target under the low SNR of traditional image processing method,and the detection accuracy is better than the traditional image processing.However,the sonar target detection method based on Faster R-CNN has to be improved in real-time detection.Therefore,in view of the above problems with Faster R-CNN,this paper attempts to apply the SSD target detection algorithm model based on regression theory to sonar target detection and improve the adaptability.Through experimental research,it is found that in the experimental environment of this paper,sonar target detection method based the SSD that can almost achieve the effect of the survivor target detection network based on Faster R-CNN,and the test speed is four times faster than the sonar target detection algorithm based on Faster R-CNN.Through the research on the key technologies of detection and recognition of these sonar target,the paper hopes to provide more ideas for the development of underwater sonar target detection and recognition systems in the future.