基于卷积神经网络的壁面清洗机器人障碍物检测识别算法  

Investigation on obstacle detection and recognition of fa ade-cleaning robot based on convolutional neural network

作　　者：郭润兰[1] 史方青 范雅琼 何智 GUO Run-lan;SHI Fang-qing;FAN Ya-qiong;HE Zhi(School of Mechanical&Electrical Engineering,Lanzhou Univ.of Tech.,Lanzhou 730050,China)

机构地区：[1]兰州理工大学机电工程学院,甘肃兰州730050

出　　处：《兰州理工大学学报》2022年第4期83-89,共7页Journal of Lanzhou University of Technology

基　　金：国家自然科学基金(51565030)。

摘　　要：结合机器人的工作原理以及卷积神经网络(CNN)在图像分类中的应用,提出了一种基于卷积神经网络的壁面障碍物检测识别算法.首先,以壁面障碍物准确识别为目标,构建壁面障碍物图像库;然后,通过对VGG-16网络简化后进行优化,得到适合壁面障碍物准确识别的卷积神经网络模型.在此基础上,设计该网络由输入层、4层卷积层、2层池化层、1层全连接层以及输出层组成,进一步利用3×3卷积核对训练样本进行卷积操作,并将所获取的特征图以2×2领域进行池化操作.重复上述操作后,通过学习获取并确定网络模型参数,得到最佳网络模型.实验结果表明,障碍物的识别准确率可达99.0%,具有良好的识别能力.Combined with the working principle of the robot and the application of convolutional neural network(CNN)of image classification,a wall obstacle detection and recognition algorithm based on CNN is proposed.Firstly,with the goal of accurate recognition of wall obstacles,the image database of wall obstacles is constructed,and then the simplified VGG-16 network is optimized to obtain a CNN model,which is suitable for accurate recognition of wall obstacles.On this basis,the network is designed to be composed of an input layer,four convolutional layers,two pooling layers,a fully-connected layer and an output layer.Further,the training samples are convolved using 3×3 convolution kernels,and the acquired feature maps are pooled in 2×2 domains.After repeating the above operations,the optimal network model is obtained by learning and the network model parameters are further determined.The experimental results show that the recognition accuracy of obstacles can reach 99.0%,which has good recognition ability.

关 键 词：清洗机器人 障碍物识别 卷积神经网络 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]