摘要
针对仓储环境下仓储机器人工作效率的问题,提出基于深度学习的卷积神经网络方法实现仓储机器人对托盘的检测。通过卷积神经网络关于物体检测方面的4种模型的比较,建立Faster RCNN模型。通过设计托盘检测模型,训练VGG16网络和自建仓储托盘数据库,实现仓储机器人对仓储环境中托盘的检测。通过对样本进行扩充,增加样本数量,再与其它3个模型进行比较。实验证明,该模型能提高仓储机器人检测托盘的准确率(mAP)。
To improve the working efficiency of forklift in the warehouse, a method for detecting warehouse pallet is put forward combined with convolution neural network of deep learning. Faster RCNN model is built by comparing four kinds of models of object detection of convolution neural network. The warehouse robot detection for storage environment is implemented by designing pallet detection model, training VGG16 network and building database of warehouse pallet. Through expanding the samples, the quantity of samples is increased and then compared with other three models. By experiment, the data speed and the mean average accuracy (mAP) of image processing are obtained.
出处
《北京信息科技大学学报(自然科学版)》
2017年第2期78-84,94,共8页
Journal of Beijing Information Science and Technology University
基金
基金项目:促进高校内涵发展--智能物流机器人研发平台(5211623100)