基于改进P-Unet模型的岩屑颗粒识别  

Identification of cuttings particle based on improved P-Unet model

作　　者：万川 王正勇[1] 何海波 滕奇志[1] 何小海[1] Wan Chuan;Wang Zhengyong;He Xiaobo;Teng Qizhi;He Xiaohai(College of Electronics and Information Engineering,Sichuan University,Chengdu 610065,China;Chengdu Xitu Technology Co.,Ltd.,Chengdu 610065,China)

机构地区：[1]四川大学电子信息学院,四川成都610065 [2]成都西图科技有限公司,四川成都610065

出　　处：《信息技术与网络安全》2020年第11期56-61,共6页Information Technology and Network Security

基　　金：国家自然科学基金(61372174)。

摘　　要：提出了一种基于改进P-Unet模型的岩屑颗粒识别方法。该方法基于Unet模型结构,运用金字塔池化模块聚合不同区域的上下文特征信息,以充分利用全局信息。改进P-Unet模型采用了残差网络ResNeXt101,在提高岩屑颗粒识别准确率前提下,减少了超参数数量。该模型采用了焦点损失函数,在一定程度上解决岩屑颗粒类别不平衡的问题,同时运用深度可分离卷积代替传统卷积,较大程度减少了网络的参数以及预测的时间。实验结果表明,改进P-Unet模型得到的识别准确率对比同类先进算法有一定的提升,对岩屑颗粒识别的结果更加准确。This paper proposes a model based on improved P-Unet cuttings particle identification method.The method is based on the Unet model structure,and uses pyramid pooling module to aggregate contextual feature information in different regions to make full use of global information.The improved P-Unet model uses the residual network ResNeXt101,which reduces the number of super parameters on the premise of improving the accuracy of debris particle recognition.In this model,the focus loss function is used to solve the problem of the unbalance of cuttings particles.At the same time,the deep separable convolution is used to replace the traditional convolution,which greatly reduces the parameters of the network and the prediction time.The experimental results show that the recognition accuracy of the improved P-Unet model is improved compared with similar advanced algorithms,and the recognition results of cuttings particles are more accurate.

关 键 词：岩屑颗粒识别 Unet网络 P-Unet网络 金字塔池化模块 深度可分离卷积 焦点损失函数 残差网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]