深度强化学习联合回归目标定位  

Union Regression Object Localization Based on Deep Reinforcement Learning

作　　者：姚红革[1] 张玮 杨浩琪 喻钧[1] YAO Hong-Ge;ZHANG Wei;YANG Hao-Qi;YU Jun(School of Computer Science and Engineering,Xi'an Technological University,Xi'an 710021)

机构地区：[1]西安工业大学计算机科学与工程学院,西安710021

出　　处：《自动化学报》2023年第5期1089-1098,共10页Acta Automatica Sinica

摘　　要：为了模拟人眼的视觉注意机制,快速、高效地搜索和定位图像目标,提出了一种基于循环神经网络(Recurrent neural network, RNN)的联合回归深度强化学习目标定位模型.该模型将历史观测信息与当前时刻的观测信息融合,并做出综合分析,以训练智能体快速定位目标,并联合回归器对智能体所定位的目标包围框进行精细调整.实验结果表明,该模型能够在少数时间步内快速、准确地定位目标.To simulate the visual attention mechanism of the human eye,search and locate image objection quickly and efficiently,this paper proposes a union regression deep reinforcement learning object localization model based on recurrent neural network(RNN),which fuses the historical observation information with the observation information at the current time,then makes a comprehensive analysis to train the agent to quickly locate the object,and combine with the regressor to fine-tune the object bounding box positioned by the agent.Experiments show that the proposed model can accurately and rapidly locate the object in a few time steps.

关 键 词：视觉注意机制 循环神经网络 深度强化学习 目标定位 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]