基于改进Mask R-CNN的堆积式袋装炸药识别方法研究  

Research on the identification method of stacked bagged explosives based on improved Mask R-CNN

作　　者：付晓鸽 左治江[1,2] 李涵 FU Xiaoge;ZUO Zhijiang;LI Han(State Key Laboratory of Percision Blasting,Jianghan University,Wuhan 430056,China;Hubei Key Laboratory of Blasting Engineering,Wuhan 430056,China)

机构地区：[1]江汉大学精细爆破国家重点实验室,湖北武汉430056 [2]爆破工程湖北省重点实验室,湖北武汉430056

出　　处：《中国测试》2024年第10期81-86,共6页China Measurement & Test

基　　金：国家重点研发计划资助项目(2021YFB2301004)。

摘　　要：针对爆破场景下,光照变化、袋装炸药紧密堆积以及袋与袋粘连导致边界模糊等问题,提出一种基于改进Mask R-CNN的堆积式袋装炸药识别方法。该文在Mask R-CNN的基础上采用空洞卷积代替普通卷积,引入“扩张率”参数,使得袋装炸药的边缘特征得到充分保留。利用Faster R-CNN网络模型、FCN网络模型、Mask R-CNN网络模型和改进Mask R-CNN网络模型对相同的数据集进行检测,对比袋装炸药边缘分割的效果。实验结果表明:该文提出基于改进Mask R-CNN的堆积式袋装炸药识别方法对袋装炸药边缘信息保存较为完整,平均准确率达到90.42%,平均速度达到0.67 s/piece,为袋装炸药装卸搬运实现更高程度的自动化提供有力的技术支撑。Aiming at the problems such as the change of light,the close accumulation of bagged explosives and the blurring of the boundary caused by the adhesion between bags in the blasting scene,a method for identifying stacked bagged explosives based on improved Mask R-CNN is proposed.In this paper,based on Mask R-CNN,cavity convolution is used instead of ordinary convolution,and the "expansion rate" parameter is introduced,so that the edge features of bagged explosives can be fully retained.Faster R-CNN network model,FCN network model,Mask R-CNN network model and improved Mask R-CNN network model are used to detect the same data set,and the effect of edge segmentation of bagged explosives is compared.The experimental results show that the method proposed in this paper based on the improved Mask R-CNN for the identification of packed explosives can preserve the edge information of packed explosives more completely,with an average accuracy rate of 90.42% and an average speed of 0.67 s/piece,which provides a strong technical support for the higher degree of automation of loading and unloading of packed explosives.

关 键 词：实例分割 Mask R-CNN 堆积式袋装炸药 神经网络 

分 类 号：TB9[一般工业技术—计量学] TP391.4[机械工程—测试计量技术及仪器] TP183[自动化与计算机技术—计算机应用技术]