摘 要：为了解决传统仓储托盘检测方法泛化性差，检测精度低的问题。设计了一种基于Faster RCNN 深度学习算法的仓储托盘检测模型，对算法模型进行了网络、数据增强处理以及特征提取方面的优化。自主拍摄仓储托盘图片并对其进行数据扩充，使用LableImage 平台进行数据标注，在ResNet 框架下进行网络训练，通过对比试验，改进后的模型性能高于其他常见目标检测模型，其准确率达到了96.5%，平均检测时间为76.9 ms，表明该方法能够满足工业生产环境中对仓储托盘的检测需求。
关键词：深度学习；仓储托盘；Faster RCNN；目标检测
DOI:10.19850/j.cnki.2096-4706.2021.02.015
中图分类号：TP183 文献标识码：A 文章编号：2096-4706（2021）02-0057-07
Storage Pallet Detection Method Based on Faster RCNN Network
ZHANG Yahui1，YANG Lin1，BAI Xue2
（1.School of Mechanical Electronic & Information Engineering，China University of Mining and Technology-Beijing，Beijing 100083，China；2.Zibo Education Service Center，Zibo 255030，China）
Abstract：In order to solve the problem that the traditional storage pallet detection method has the low generalization and low detection accuracy. A storage pallet detection model based on the Faster RCNN deep learning algorithm is designed，and the algorithm model is optimized in the aspects of network，data enhancement processing and feature extraction. Which can achieve independently take pictures of storage pallets and make data augmentation on them，use the LableImage platform for data annotation，and conduct network training under the ResNet framework. Through the comparative experiments，the improved model performance is higher than other common target detection models，and its accuracy rate reaches at 96.5 %，the average detection time is 76.9 ms，the results show that the method can meet the detection requirements for storage pallets in industrial production environment.
Keywords：deep learning；storage pallet；Faster RCNN；target detection
参考文献：
[1] 胡迟. 制造业转型升级：“十二五”成效评估与“十三五”发展对策 [J]. 经济研究参考，2016（49）：3-27.
[2] 徐翔. 仓储行业正经历一场智能化大变革 [J]. 中国储运，2020（10）：50-51.
[3] LYU Z J，ZHAO P C，LU Q，et al. Prediction of the Bending Strength of Boltless Steel Connections in Storage Pallet Racks：An Integrated Experimental-FEM-SVM Methodology [J/OL].Advances in Civil Engineering，2020：[2020-10-22].https://doi.org/10.1155/2020/5109204.
[4] PARKHI O M，VEDALDI A，ZISSERMAN A. Deep Face Recognition [C]//Proceedings of the British Machine Vision Conference. Swansea：BMVA Press，2015：41.1-41.12.
[5] LIU W，ANGUELOV D，ERHAN D，et al. SSD：Single Shot MultiBox Detector [J/OL].arXiv：1512.02325 [cs.CV].（2016-12-29）.https://arxiv.org/abs/1512.02325.
[6] REN S Q，HE K M，GIRSHICK R，et al. Faster R-CNN：Towards Real-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2017，39（6）：1137-1149.
[7] GIRSHICK R，DONAHUE J，DARRELL T，et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [C]//2014 IEEE Conference on Computer Vision and Pattern Recognition.Columbus：IEEE，2014：580-587.
[8] 卢宏涛，张秦川. 深度卷积神经网络在计算机视觉中的应用研究综述 [J]. 数据采集与处理，2016，31（1）：1-17.
[9] ERGUN H，SERT M. Fusing Deep Convolutional Networks for Large Scale Visual Concept Classification [C]//2016 IEEE Second International Conference on Multimedia Big Data（BigMM）.Taipei：IEEE，2016：210-213.
作者简介：张亚辉（1994—），男，汉族，河南平顶山人，硕士研究生在读，研究方向：数据挖掘与人工智能。