多光谱分离算法在目标识别中的研究

针对野外工作时机械扫描式的光学系统抗震性差、目标识别率低、实时性差等问题,设计了采用多光谱分离算法实现非扫描目标识别遥感系统。采用非扫描的M-Z干涉具提供空间光程差,由红外CCD采集干涉条纹信息,经CUP处理得到混合光谱,结合可见光视频图像提供的坐标系实现识别目标。其中采用遗传算法优化选择特征波长,然后由粗糙集分类提取未知目标谱的属性,取前1/3可信度的相应属性反演待测目标种类,相比传统算法减少约9倍的运算量。在不同天气、不同背景条件下做实验,得到系统在各种情况下的探测极限及识别概率。由实验数据可知,采用遗传算法和粗糙集分类相结合的多光谱分离算法可以快速、有效地识别未知目标的种类。

Research on target identification by multi-spectrum separation algorithm

In view of problems such as field poor shock resistance, low target identification rate, low real-time and so on in mechanical scanning optical system, a non-scanning target identification remote sensing system was designed using the multi-spectrum separation algorithm. Using the non-scanning M-Z interferometer to provide a space optical path difference, interference fringes were collected by infrared CCD detector. After CUP processing the system obtains the mix spectrum information, achieves target identification by the coordinate system combined with visible light video image, and the coordinate system, which the union visible light video image provides, achieves the target discrimination The genetic algorithm was used to optimize characteristic wavelengths, and then by the rough collection classification the unknown target spectrum's attribute was extracted. Taking first 1/3 confidence level of the corresponding attribute the testing target type was deduced, and compared with the traditional algorithm the amount of computing was reduced by about nine times. Experiment was done under different weather and different background conditions, so detection limits and identification probabilities of the system under different conditions were obtained. The experimental data showed that the genetic algorithm and rough set classification combined with multi-spectral separation algorithm can quickly and efficiently identify the unknown object types.