传统的合成孔径雷达(SAR)目标识别往往采用批量学习的方法, 但是在现实应用中, 系统的训练数据并不能一次性全部获得, 当有新的训练样本到来时, 采用批量学习方法需要重新训练整个系统。为解决这个问题, 将增量学习算法——正则在线序贯式极限学习机(ROSELM)应用到SAR目标识别中, 并且采用粒子群算法优化ROSELM的初始权值以提高其稳定性和识别率。实验结果表明, 该算法在新的SAR目标样本获得时只需要通过更新输出权值即可完成系统的更新, 无需重新训练, 且速度极快、识别率高, 可以作为SAR目标识别系统在线更新的良好选择。

Target recognition of conventional Synthetic Aperture Radar (SAR) usually adopts the batch learning method.However, in practical application, the training data of the system cant be obtained all at one time.When new training sample arrives, the whole system needs to be retrained when using the method of batch learning.In order to solve this problem, ROSELM, an incremental learning algorithm, is applied to SAR target recognition, and Particle Swarm Optimization(PSO) algorithm is used to optimize the initial weight of ROSELM to improve its stability and recognition rate.The experimental results show that:1) When new SAR target samples are obtained, the system updating can be implemented simply by updating the output weights without re-training;2) The algorithm is very fast and has a high recognition rate, which is a good choice for online updating of SAR target recognition system.