Research On Small Target Detection In Sea Based On SVM

The research on small target detection in sea has been a hot point in the field of radarsignal processing, playing an important role on engineering applications. With the Chineseaircraft carrier success on the first trial, the research on small target detection in seacontinues to heat up in the country. Since the signal of small target is often lost in sea clutter,it is difficult to detect directly. Research on sea clutter is a necessary way to detect smalltargets indirectly. However, the result of traditional statistical probability model of seaclutter is unsatisfactory in the case of low SNR. Neural network as a representative oflearning machines has an extraordinary performance in sea clutter prediction. However, itsinherent defects limit its development, such as hidden node selection, easy to fall into localminimum point.Based on the background, chaotic model of sea clutter is established, and sea clutter ispredicted by using support vector machine. and the small target is detected. The main job ofthis subject is as follow:(1) The chaotic dynamic characteristic of sea clutter is researched, and through thesimulation, the fact is verified that the sea clutter has chaos characteristics. The model ofchaos is established and the sea clutter can be short-term predicted;(2) Support vector machine as a learning machine is trained by the measured data ofradar. The sea clutter signal is predicted through simulation, and according to the differenceof relative mean square error energy, the small target is detected;(3) The noise of sea clutter is reduced by wavelet transform. The experiment reducesmost of noise mixed in the sea clutter through different wavelet base, different thresholdand different way to threshold. Most of noise is removed;(4) The best matching parameters of support vector machines are global searched byparticle swarm algorithm. The result of simulation shows that, in larger search area, withgood sea clutter prediction effect, particle swarm optimization algorithm can quickly searchthe best matching parameters of support vector machines.