Ship Target Classification And Line Spectrum Detection Based On Artificial Intelligence

With the continuous development of artificial intelligence technology in recent years,significant progress has been made in the application of AI technology in various fields,such as facial recognition,vehicle identification,speech processing,medical image recognition,and more.In the field of target detection and recognition of ship radiation noise,the research focus has shifted from traditional feature-based fuzzy matching to AI-based target detection and recognition.Deep learning,a new branch of AI technology,relies heavily on big data support,which happens to be a weak point in ship radiation noise research.Therefore,the key challenge in this field is how to achieve better detection and recognition results using deep learning techniques with limited sample data.This thesis focuses on ship radiation target recognition and line spectrum detection and presents the following aspects of work:1.Simulation modeling and feature extraction denoising of ship radiation noise.This thesis primarily utilizes LOFAR spectrogram and DEMON spectrogram analysis as the main feature spectrograms of ship radiation noise.LOFAR spectrogram employs Welch’s method as the main power spectrum analysis technique,while three different filtering algorithms are used to estimate the continuous spectrum background interference in DEMON spectrogram features,achieving the separation of DEMON spectrogram’s continuous spectrum and line spectrum components.By modeling and programming,simulated data is generated to compare the performance of the three filtering algorithms on DEMON spectrogram processing and provide support for simulated data and datasets in subsequent sections.2.Research on fusion classification and recognition models combining unsupervised and supervised learning.This thesis primarily focuses on unsupervised learning models such as Principal Component Analysis(PCA),Autoencoder(AE),and Deep Belief Network(DBN),followed by a supervised learning Multi-Layer Perceptron(MLP)classification neural network using the Backpropagation(BP)algorithm.Pretrained unsupervised learning models help reduce the risk of overfitting in the classification and recognition network.First,different feature inputs and fusion classification models are analyzed through simulated data experiments to evaluate their classification and recognition performance.Then,the effectiveness of the proposed fusion classification models is validated using actual data from two types of ships.3.Research on line spectrum detection methods based on image recognition.This thesis proposes two line spectrum detection schemes based on traditional image processing methods and deep learning-based image processing methods.The first method is a three-threshold line spectrum detection method,which is compared with the Canny operator’s line spectrum detection method in simulated experiments.The experimental results show that the three-threshold line spectrum detection algorithm performs better under interference-free conditions.However,when white noise interference is present in the signal,the three-threshold line spectrum detection algorithm may suffer from weak line spectrum omission and false line spectrum detection.The second method is the SAM+YOLOv5+EIo U model for line spectrum detection.Simulation results demonstrate that this method performs well under interferencefree conditions and can balance the line spectrum detection rate and precision by adjusting the confidence threshold of the model when white noise interference is present.Finally,experimental results with actual data indicate that the SAM+YOLOv5+EIo U model has the potential for auxiliary target presence/absence detection.