Research On Detection And Recognition Of Underwater Target Based On Network And Data Fusion

Recognition technique on underwater targets has important realistic meaning and civil value and it is the emphases of learners studying all over the world. Most countries in the world pay much attention to it and a lot of achievements have been obtained. The learners of our country have contributed to this. Based on this the paper mainly analyzes many means of the underwater target feature's extraction and research the recognition means based on the article neutral network. And the results of the experiment show that the spectrum-feature is effective.Following are the primary contributions of this paper.1. The theory of wavelet-transformed is studied systematically. The basic principle, means and effects of the wavelet transformation were analyzed. Every factor which impacts the effect of signal denoise by wavelet-transformed was summarized by means of simulation. Based on this the ship noise was denoised by wavelet-transformed and good results were obtained. Then the features of signal noise were extracted by wavelet-transformed. The classification experiment for three different classes of targets was done by the features and BP network.2. The basic characters and effects of higher-order statistics were discussed and analyzed in this paper. The 11/2 and 21/2 dimensional spectrum were studied mainly. And the feature of 11/2 and 21/2 dimensional spectrum were extracted. The classification experiment for three different classes of targets was done by BPnetwork and the mended BP network.3. The meaning, the basic character and academic model of power spectrum on underwater signal were discussed. The line-spectrum was extracted and optimized.4. The structure and type of artificial network, including BP artificial network which is often used were analyzed. And the structure, the rules of study and how to design and training BP artificial network were analyzed in this paper.5. The shorts of recognition by the single sensor and one feature were researched, which showed the necessity of using multi-artificial network. A mean of max probability to judge the result of feature fusion was presented. The result of fusion was good and the classification rate was improved.