结合变异粒子群和字典学习的遥感影像去噪

发布时间：2018-03-08 12:14

本文选题：变异粒子群　切入点：在线字典学习　出处：《计算机工程与科学》2017年09期 　论文类型：期刊论文

【摘要】：针对在线字典学习需将所有字典原子全部更新、优化方向难以进行估算等原因造成精度下降的不足,提出基于变异粒子群优化的在线字典学习算法。算法基于ODL的基础,在字典学习的迭代过程中对梯度下降函数进行优化。首先选出特殊字典原子,利用各个字典原子之间关系,线性表征当前选出的原子,以线性系数作为粒子群中的粒子位置。然后将基于变异粒子群的原子更新模式引入字典学习,利用变异粒子群优化算法进行粒子的适应度淘汰,选择更适合的粒子进行下一轮的字典更新。此外,利用中间变量将历史参考数据引入变异粒子群模型以引导其优化方向,提高字典的准确性和有效性。利用高分一号遥感影像进行实验,实验结果表明该算法优于同类方法,有更好的噪音抑制效果,同时也提高了大规模的遥感图像处理性能。
[Abstract]:In order to solve the problem that all the dictionary atoms need to be updated and the optimization direction is difficult to estimate, an online dictionary learning algorithm based on mutation particle swarm optimization (VPSO) is proposed. The algorithm is based on ODL. In the iterative process of dictionary learning, the gradient descent function is optimized. Firstly, special dictionary atoms are selected, and the current selected atoms are represented linearly by using the relationship between each dictionary atom. The linear coefficient is used as the particle position in the particle swarm, and then the atomic renewal model based on the variable particle swarm is introduced into the dictionary learning, and the particle fitness is eliminated by using the mutation particle swarm optimization algorithm. Select more suitable particles for the next dictionary update. In addition, historical reference data are introduced into the variable particle swarm model to guide the optimization direction. The accuracy and effectiveness of the dictionary are improved. The experimental results show that the proposed algorithm is superior to the similar methods and has better noise suppression effect. At the same time, it also improves the performance of large-scale remote sensing image processing.
【作者单位】：中国科学院遥感与数字地球研究所;中国科学院大学;
【基金】：国家科技支撑计划(2015BAJ02B00)
【分类号】：TP751

【相似文献】