智能机器狗巡线数据预处理方法探析

Analysis of data preprocessing method for intelligent robot dog patrol

智能机器狗作为人工智能端侧设备在生产生活中有广泛的应用场景。基于深度学习模型的智能巡线是机器狗的重要功能,训练巡线模型需要准备丰富的数据集,同时要求结合实际场景对数据集进行合理的预处理。首先,结合机器狗单向巡线功能应用场景指出了巡线图像数据集预处理时应注意的旋转、翻转问题,避免机器狗偏离航线问题;其次,结合图像设备采集数据质量较差的情况,指出图像增强的必要处理方法集,以及在硬件设备性能限制的情况下推荐图像增强方法;最后,根据ResNet50模型训练部署结果,对巡线图像数据集预处理存在的问题和改进空间作了分析。

As an artificial intelligence end device,intelligent robot dog has a wide range of application scenarios in production and life.The intelligent line patrol based on deep learning models is an important function of the robot dog.Training the line patrol model requires preparing a rich dataset,and also requires reasonable preprocessing of the dataset based on actual scenarios.To begin with,this article points out the rotation and flipping issues that should be paid attention to during the preprocessing of the line patrol image dataset,based on the application scenario of the machine dog one-way line patrol function,in order to avoid the problem of the machine dog deviating from the route;Secondly,considering the poor quality of data collected by image devices,this article also points out the necessary set of processing methods for image enhancement,as well as the recommended image enhancement methods for hardware device performance limitations;Finally,based on the training and deployment results of the ResNet50 model,this article analyzes the problems and improvement space in the preprocessing of the patrol image dataset.