基于KPCA联合并联抑制神经网络变换的红外目标识别算法

Infrared target recognition based on joint KPCA shunting inhibition neural network transformation

为了提高红外目标的识别性能,提出了一种KPCA联合并联抑制神经网络变换.该联合神经网络变换集成了KPCA的KHA学习机制与神经网络误差反传机制,使得KPCA与GSN分类器有机地结合起来,通过监督学习的方式引入类别信息,能够在实现数据有效降维的同时,优化主元特征的提取,从而提高算法的分类识别性能.针对典型红外军用车辆图像,采用联合算法与传统算法分别进行对比实验.实验结果表明,算法在优化特征同时,提高了目标识别性能.

A novel joint KPCA （kernel principal component analysis） shunting inhibition neural network transformation is presented to improve the recognition capability of infrared target. The transformation integrates KPCA and GSN（ generalized shunting neuron） classifier by combining together the learning mechanism of kernel Hebbian algorithm （KHA） based on neural network and a back propagation learning algorithm. Then the interclass information is introduced through the supervised learning manner, and better dimensionality reduction and a higher degree of discriminability can be achieved. The joint method was applied to typical infrared military vehicles in comparison with several traditional methods. Experimental results show that the method can offer more powerful target recognition with optimized feature extraction.