Research On Underwater Metal Target Detection Method

Underwater target detection and recognition is a core technology in the research of underwater unmanned detection methods,and is of great significance for subsequent battlefield situation assessment,threat estimation,and attack strategy formulation.Optical detection can achieve good target classification results,but is suitable for short-range detection.Acoustic detection can discover targets at long distances and output distance information,but it is difficult to construct the target contour.Magnetic detection is only suitable for close-range detection and can assist other sensors in conducting magnetic detection.Therefore,relying solely on a single sensor for underwater detection inevitably has certain limitations and cannot accurately and completely obtain target information.Target detection methods based on multi-sensor data fusion can make full use of multi-source information related to system recognition function,and combine and reason to make up for each other’s strengths and weaknesses,so that the system can obtain more comprehensive and accurate observations of target characteristics,significantly improving the system’s recognition rate,robustness,universality,and anti-false target interference performance.Therefore,this paper uses three types of sensors,including optical sensors,acoustic sensors,and magnetic sensors,to jointly construct a submarine detection system.The paper studies the detection and recognition methods of underwater metal targets and multi-sensor decision fusion technology,and verifies the effectiveness of relevant algorithms through experiments.The main work and innovative achievements are as follows:(1)Convolutional Neural Network(CNN)algorithm is applied to the processing of underwater optical images,and combined with the characteristics of underwater optical images,Through K-Singular Value Decomposition(K-SVD)algorithm and improved Contrast Limited Adaptive Histogram Equalization,CLAHE(CLAHE)algorithm enhances the contrast between the target and the background for preprocessing,to obtain better quality images,and to solve the problem of noise and inhomogeneity in the image.CNN is used to recognize the processed image.The algorithm can be built into a lightweight CNN architecture through Tensor Flow,which is convenient for subsequent transplantation.(2)A new quantum-classical hybrid classification algorithm is proposed to address the problems of large sonar data volumes,insufficient computing resources,and slow processing speeds.This algorithm utilizes quantum algorithms for dimensionality reduction and classical algorithms for data classification.Using two types of underwater target sonar data sets as examples,the hybrid algorithm improves the classification accuracy from 0.772 to 0.821 compared to classical algorithms,even considering the classical quantum data readout.The hybrid algorithm also has polynomial acceleration in dimensionality reduction compared to classical methods,greatly improving classification speed.In the future,it can be used for fast processing of data from detectors and accurate classification.(3)Based on the STM32 development platform,a hardware circuit for ultrasonic sensor transmission and reception was designed,which includes four modules: signal generation circuit,filtering circuit,amplification circuit,and comparison circuit.Several typical time delay estimation algorithms were analyzed,and several algorithms were compared in terms of implementation method,time delay estimation accuracy,algorithm complexity,and real-time performance.The cross-correlation time delay estimation method with better comprehensive performance was selected and applied to the ultrasonic ranging module constructed in this paper.The experimental results show that the sensor can meet the detection requirements,and the ranging error is within 6mm.(4)To address the problem of the sensor being susceptible to background magnetic fields,interference noise,and weak signals from underwater magnetic targets,a magnetic detection module with an excitation magnetic field was designed.This method increases the sensitivity of the sensor to magnetic targets in the environment by adding a magnetic induction coil to produce an excitation magnetic field,and improves the anti-interference ability.Based on the characteristics of the amplitude change of the target signal,a threshold detection algorithm based on the minimum information entropy(MIE)filter was proposed.The detection performance of the algorithm was evaluated using noise+signal patterns.The simulation results show that under a constant excitation magnetic field,when the closest point of approach(CPA)is 2 meters,the target signal detection probability reaches about 80%,and when the CPA is 4 meters,the target signal detection probability decreases to about 70%.When there is noise in the environment,the recognition rate of magnetic sources is better than that without magnetic sources as the detection distance increases.(5)Aiming at the problem of evidence conflict in the decision-level fusion of underwater metal target detection using the optical-acoustic-magnetic multi-sensor,a modified evidence synthesis algorithm based on the Dempster-Shafer(DS)evidence theory is proposed.The deficiencies of the DS evidence theory and the reasons for the conflict are thoroughly analyzed,and a method based on a new conflict coefficient and DS combination is proposed to allocate the evidence conflict according to the degree of conflict of the evidence on different targets.The improved evidence theory is used to conduct decision-level fusion experiments on the data results of the optical-acoustic-magnetic sensors.The experimental results show that the improved algorithm effectively improves the accuracy of underwater metal target detection and identification.This paper deeply investigates and discusses the key technologies of underwater metal target detection based on multi-sensor data fusion.More comprehensive target information is obtained through the cooperation of three types of sensors,and the effectiveness and feasibility of the relevant work are verified by taking minesweepers as an example,providing accurate guidance for the motion planning and decision control of subsequent underwater vehicles.The proposed method can also be used in other military target reconnaissance,deep-sea metal mineral exploration,shipwreck search,etc.,laying a theoretical foundation for the development of underwater metal target detection systems.