基于BP神经网络提高伪装目标识别概率的研究

采用静态迈克尔逊干涉仪对待测目标进行光谱识别,在空间干涉长度不变的条件下,应用BP神经网络算法对混合光谱分离过程进行优化,从而达到提高伪装目标识别概率的目的。由干涉仪及线阵CCD记录视场内所有位置上的光谱信息,构成混合光谱数据集合,以已知材料的标准吸收光谱作为隐含层的规则依据,将BP神经网络应用于混合光谱的分离。实验采用不同距离、不同背景组合的混合光谱作为初始数据,以1.5 m×1.5 m钢板做成四种待测目标,由静态迈克尔逊干涉仪得到混合光谱,BP神经网络算法与传统光谱吸收算法对无伪装目标的识别率都在90%以上,对具有伪装效果的待测目标识别概率分别为75.5%和31.7%,所以采用BP神经网络可有效地提高伪装目标的识别概率。

Improvement of the recognition probability about camouflage target based on BP neural network

Using static Michelson interferometer to get the spectrum information of measurement targets for spectrum identification, under the condition that the interference length is constant, the system can be optimized by BP neural network algorithm for the mixed spectral separation process. Thereby it can realize improving the recognition probability of camouflage target. Collecting the spectrum information in field of view (FOV) by the interferometer and linear array CCD detector, composing the set of mixed spectrum data, with known absorption spectrum of the material as a hidden layer of rules, it used BP neural network to separate the mixed spectrum data. Experiment with different distances, different combinations of mixed background spectrum as the initial data, using steel target (size: 1.5 m x 1.5 m) made of four kinds, the recognition probability of non-camouflage target is about 90% by BP neural network algorithm or the traditional algorithm, while the recognition probability of camouflage target is 75.5% with BP, better than 31.7% with the traditional, so it can effectively improve the recognition probability of camouflage target.