摘要
在机器人目标识别过程中,由于目标具有无序性的特征,同时受到环境复杂性、数据量庞大、噪声等因素的影响,导致目标识别精度受到严重影响。为了满足智能机器人对无序目标的识别需求,提出一种应用强化学习的智能机器人目标无序识别方法。通过双边滤波将采集的目标图像,并将其分解为低照度图像和反射图像,分别使用不同策略压缩照度图像和增强反射图像,将两部分图像合并形成新图像。将深度学习技术和强化学习有效结合,完成增强处理的目标图像输入到深度强化学习中训练,实现智能机器人目标无序识别。实验结果表明,所提方法可以有效改善目标图像质量,同时获取高效率以及高精度的智能机器人目标无序识别结果。
In the process of robot target recognition,the accuracy of target recognition is severely affected due to the disorderly characteristics of the target and the influence of factors such as environmental complexity,large data volume,and noise.In order to meet the recognition needs of intelligent robots for disordered targets,a reinforcement learning based method for disordered target recognition of intelligent robots is proposed.Through Bilateral filter,the collected target image is decomposed into low illumination image and reflection image,and different strategies are used to compress the illumination image and enhance the reflection image respectively,and the two images are combined to form a new image.Effectively combining deep learning technology with reinforcement learning,the target image that completes enhancement processing is input into deep reinforcement learning for training,achieving intelligent robot target disorderly recognition.The experimental results show that the proposed method can effectively improve the quality of the target image,while obtaining efficient and high-precision intelligent robot target disorder recognition results.
作者
徐楚原
何健
XU Chu-yuan;HE Jian(Melbourne School of Engineering and IT,The University of Melbourne,Melbourne VIC3010,Australia;School of Mathematics and Computer Science,Wuhan Polytechnic University,Wuhan Hubei 430023,China)
出处
《计算机仿真》
北大核心
2023年第10期440-444,共5页
Computer Simulation
关键词
强化学习
智能机器人
目标无序
双边滤波
Reinforcement learning
Intelligent robots
Target disorder
Bilateral filter
