基于改进DeepLabV3+的沥青成品料装车状态自动监测  

Asphalt product loading status automatic monitoring based on improved DeepLabV3+

作　　者：李许峰 王向前 王耀耀 夏晓华 LI Xufeng;WANG Xiangqian;WANG Yaoyao;XIA Xiaohua(Shaanxi Motorway Mechanization Engineering Co.,Ltd.,Xi’an 710038,China;School of Construction Machinery,Chang’an University,Xi’an 710064,China)

机构地区：[1]陕西高速机械化工程有限公司,陕西西安710038 [2]长安大学工程机械学院,陕西西安710064

出　　处：《现代电子技术》2023年第17期117-121,共5页Modern Electronics Technique

摘　　要：当前沥青拌合站成品料装车系统仍旧依赖人工观察对装车状态进行监测,存在自动化程度低、作业效率不高等问题。为实现沥青拌合站装车的自动化监测,提出一种基于语义分割的机器视觉自动监测方法,使用ShuffleNetV2替代原DeepLabV3+的骨干网络Xception,构建出轻量级的语义分割网络S-DeepLabV3+,利用S-DeepLabV3+将场景分割为车厢、料堆、落料和背景4个区域,然后利用落料语义所占百分比判断料门的落料状态,利用落料区域料堆顶点到车厢上边沿的距离判断落料区域的料堆状态。实验结果表明,与原DeepLabV3+相比,S-DeepLabV3+在保证图像分割准确率的前提下,速度提升了16.23%,其监测结果与人工监测判别结果一致,准确率可达到100%,可以满足沥青拌合站成品料装车状态监测的应用要求,为沥青拌合站实现自动化装车奠定了基础。The loading status monitoring of the current finished material loading system of asphalt mixing station still relies on manual observation,which leads to low automation and operating efficiency.Therefore,a machine vision automatic monitoring method based on semantic segmentation is proposed to realize automatic monitoring of loading of asphalt mixing station.The ShuffleNetV2 is used to replace backbone network Xception of the original DeepLabV3+,and a lightweight semantic segmentation network S⁃DeepLabV3+is constructed.The S⁃DeepLabV3+is used to divide the scene into four regions,named carriage,stack,blanking and background.And then,the blanking state is judged by the percentage of blanking semantics,and the stack state of the blanking area is judged by the distance from the stack vertex to the upper edge of the carriage.The experimental results show that,in comparison with the original DeepLabV3+,S⁃DeepLabV3+can increase the speed by 16.23%under the premise of ensuring the accuracy rate of image segmentation.The monitoring results are consistent with the manual discrimination results,and the accuracy rate can reach 100%.Therefore,it can meet the application requirements of monitoring the loading state of finished materials of asphalt mixing station and lay a foundation for realization of automatic loading of asphalt mixing station.

关 键 词：装车状态监测 沥青拌合站 散料装车 料堆识别 机器视觉 DeepLabV3+ 语义分割 自动化 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP277[电子电信—信息与通信工程]