摘 要

隨著科學(xué)技術(shù)的發(fā)展，以及國(guó)防安全的需要，對(duì)于水下目標(biāo)的識(shí)別已經(jīng)變得越來(lái)越重要。水下目標(biāo)識(shí)別是水聲裝備發(fā)展的三項(xiàng)關(guān)鍵技術(shù)（探測(cè)、定位、識(shí)別）之一，是探測(cè)系統(tǒng)智能化的重要標(biāo)志，同時(shí)也是聲納信息理論中急待解決的難題。開(kāi)展該領(lǐng)域的研究具有極其重要的現(xiàn)實(shí)意義與軍事價(jià)值。
水下目標(biāo)識(shí)別分為主動(dòng)識(shí)別和被動(dòng)識(shí)別兩種，本文研究的是被動(dòng)識(shí)別技術(shù)。它是將被動(dòng)聲納接收的水下目標(biāo)噪聲信號(hào)先進(jìn)行特征提取，提取出能夠反映目標(biāo)特征的特征向量，然后設(shè)計(jì)一個(gè)目標(biāo)分類器，最后將提取出的能夠反映目標(biāo)本質(zhì)特性的特征向量送入目標(biāo)分類器進(jìn)行分類識(shí)別。
在特征提取階段，本文將采集的水下目標(biāo)的信號(hào)進(jìn)行快速傅里葉變換（FFT）得到信號(hào)的功率譜，然后對(duì)功率譜進(jìn)行特征提取，其中最主要的特征提取方法包括連續(xù)譜特征提取、線譜特征提取、調(diào)制連續(xù)譜特征提取、調(diào)制線譜特征提取，這樣就可以得到信號(hào)的基于不同特征提取方法的特征向量。
在得到目標(biāo)的特征向量后，首先設(shè)計(jì)一個(gè)自適應(yīng)遺傳BP神經(jīng)網(wǎng)絡(luò)分類器對(duì)目標(biāo)進(jìn)行分類處理，經(jīng)仿真實(shí)驗(yàn)表明該特征分類器能夠有效地對(duì)水下目標(biāo)信號(hào)進(jìn)行識(shí)別，其識(shí)別率達(dá)到了85%以上。為了體現(xiàn)基于水下傳感器網(wǎng)絡(luò)的目標(biāo)識(shí)別，本文又采用了基于D-S證據(jù)理論的方法對(duì)目標(biāo)進(jìn)行融合識(shí)別，其過(guò)程為：首先訓(xùn)練一個(gè)BP神經(jīng)網(wǎng)絡(luò)，然后把上文所介紹的水下目標(biāo)信號(hào)的各個(gè)特征向量輸入訓(xùn)練好的BP神經(jīng)網(wǎng)絡(luò)，這樣BP神經(jīng)網(wǎng)絡(luò)輸出的就是D-S證據(jù)理論所要得到的基本概率賦值，然后利用該基本概率賦值對(duì)目標(biāo)進(jìn)行D-S融合識(shí)別，經(jīng)仿真實(shí)驗(yàn)表明該融合算法，識(shí)別率達(dá)到90%以上，目標(biāo)識(shí)別的精度明顯升高。
在實(shí)驗(yàn)室現(xiàn)有條件下，本次試驗(yàn)通過(guò)布置在水槽中的一些傳感器節(jié)點(diǎn)來(lái)模擬水下傳感器網(wǎng)絡(luò)。首先節(jié)點(diǎn)將采集的水下目標(biāo)的特征數(shù)據(jù)發(fā)送給網(wǎng)關(guān)，網(wǎng)關(guān)再通過(guò)串口將數(shù)據(jù)傳送到網(wǎng)絡(luò)控制系統(tǒng)的數(shù)據(jù)庫(kù)中；然后在數(shù)據(jù)庫(kù)中通過(guò)調(diào)用matlab程序完成對(duì)目標(biāo)的分類識(shí)別；最后利用嵌入式web實(shí)現(xiàn)對(duì)目標(biāo)識(shí)別結(jié)果的遠(yuǎn)程監(jiān)測(cè)。

關(guān)鍵字： 目標(biāo)識(shí)別；特征提?。簧窠?jīng)網(wǎng)絡(luò)分類器；遺傳算法；D-S融合；








Abstract
With the development of science and technology,with the needs for national security,it has become more and more important for us to identify the underwater target. Underwater target recognition is one of the three key technical （exploration，orientation，recognition） in the development of acoustic equipment. It is an important symbol of the intelligentized exploration system and is always one of the difficult problems，which are urgent to be resolved in the sonar information processing theory. Developing the research in this field has the most important practical meaning and martial value.
The recognition of underwater targets include active and passive identification.The passive recognition is our work.Firstly ,we extract the feature of the radiated noise form underwater targets.secondly designed a classifier，finally，in order to identify the underwater target ，we sent the eigenvectors which can reflect the characteristics of the underwater target to the classifier .
In the stage of feature extraction,in order to get the power spectrum of the signal , people FFT(Fast Fourier transform)the signal form underwater targets.In the process of feature extraction of power spectral, the main methods we used are included continuous spectrum、line spectrum、modulated continuous spectrum、modulated line spectrum feature extraction.In this case ,we can get eigenvectors based on different methods of feature extraction.
After getting the feature vectors of the targets，in order to identify the target，we designed a classifier based on genetic and BP neural network.After the simulation ，it can be seen that the classifier identify the underwater target effectively,whose recognition accuracy is 85% or more. In order to show the target identification based on underwater sensor networks ,we use a method based on D-S evidence theory to recognize the targets. The process is：firstly，the BP neural network must been trained .secondly，putting the feature vector which is described above into the trained BP neural network，in this way, the output form BP neural network is the basic probability assignment which is needed for D-S evidence theory.Finally,we recognized the target with the method of D-S theory. After the simulation, it can be seen that this method can improve the recognition accuracy Significantly, whose recognition accuracy is 90% or more.
Under the existing conditions of the laboratory,we simulated the underwater sensor networks by arranging in a number of sensor nodes. Firstly ,the ensor nodes sent the collected dates to the gateway，the gateway transfer the data to the datebase of the network control system with the serial port. Secondly ,the target is identified in the database by calling matlab program,Finally，achieving the remote monitoring of recognition results with the embedded web.

Key words: target recognition ；feature extraction；neural network classifier；genetic algorithm；D-S fusion；


目 錄
摘 要 …………………………………………………………………………………………………………………………… I
Abstract………………………………………………………………………………………………………………………… III
第1章 緒論……………………………………………………………………..