深度强化学习的图像特征高效分类方法仿真

Simulation of Image Feature Efficient Classification Method for Deep Reinforcement Learning

针对现有的图像特征分类方法收敛性差,分类无法满足日益增加的网络需求的现状,本文提出了一种基于深度强化学习的图像特征分类方法。通过对目标图像特征区域进行复域Contourlet分解,过滤处理分解结果,从而可以将目标图像子带系数矩阵提取出来,求取系数矩阵的相关特征。采取深度学习网络,使所选图像的特征向量直接经过已训练的层状网络深度模型,完成图像特征分类。实验结果表明,所提方法的误识率比现有方法明显降低,收敛速度明显提升。改进方法比传统方法更具优势,能够满足图像特征分类智能化处理的需要。

In this article,an efficient classification method of image feature based on deep reinforcement learning was put forward.The complex domain Contourlet transform method was used to perform complex domain Contourlet decomposition on the feature region of target image,and the decomposition result was input into the filter bank.Through the filtering processing,the subband coefficient matrix of target image was extracted to find relevant features of coefficient matrix.Meanwhile,these fusion results were taken as feature vectors of target image.Finally,the deep learning network was adopted,so that the feature vector of selected image could directly pass through the trained lay-ered network depth model.Thus,the image feature classification was completed.Simulation results show that,com-pared with the existing method,the false recognition rate of proposed method is reduced obviously.The convergence speed is significantly improved.The improved method is more advantageous than the traditional method,which can meet the intelligent classification of image feature.