Lightweight mask R-CNN for instance segmentation and particle physical property analysis in multiphase flow

IF 4.5 2区工程技术 Q2 ENGINEERING, CHEMICAL

Powder Technology Pub Date : 2024-10-16 DOI:10.1016/j.powtec.2024.120366

Mingxiang He , Kexin He , Qingshan Huang , Hang Xiao , Haidong Zhang , Guan Li , Aqiang Chen

{"title":"Lightweight mask R-CNN for instance segmentation and particle physical property analysis in multiphase flow","authors":"Mingxiang He , Kexin He , Qingshan Huang , Hang Xiao , Haidong Zhang , Guan Li , Aqiang Chen","doi":"10.1016/j.powtec.2024.120366","DOIUrl":null,"url":null,"abstract":"<div><div>A lightweight Mask R-CNN instance segmentation model was developed here to analyze particle size and shape accurately and quickly. Firstly, a hybrid Depthwise Dilated Convolutional Network (DDNet) is proposed, and the feature pyramid layers and the shared convolutional layers of the region proposal network are simplified, reducing the model complexity while ensuring robust feature extraction capabilities. Then, segmentation accuracy is significantly improved without sacrificing computational speed and performance by introducing the Dice loss function and clustering algorithm. Experimental results show that the model parameters are significantly reduced by 49.46%, and the segmentation speed increases from 2.15 FPS (frames per second) to 5.88 FPS. Meanwhile, the segmentation accuracy (AP50) increased from 90.56% to 91.21%. In addition, it was proven that the particle size distribution and shape could be analyzed accurately and rapidly with the proposed model, providing essential information for multiphase flow process optimization and equipment design in industrial applications.</div></div>","PeriodicalId":407,"journal":{"name":"Powder Technology","volume":"449 ","pages":"Article 120366"},"PeriodicalIF":4.5000,"publicationDate":"2024-10-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Powder Technology","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0032591024010106","RegionNum":2,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ENGINEERING, CHEMICAL","Score":null,"Total":0}

引用次数: 0

Abstract

A lightweight Mask R-CNN instance segmentation model was developed here to analyze particle size and shape accurately and quickly. Firstly, a hybrid Depthwise Dilated Convolutional Network (DDNet) is proposed, and the feature pyramid layers and the shared convolutional layers of the region proposal network are simplified, reducing the model complexity while ensuring robust feature extraction capabilities. Then, segmentation accuracy is significantly improved without sacrificing computational speed and performance by introducing the Dice loss function and clustering algorithm. Experimental results show that the model parameters are significantly reduced by 49.46%, and the segmentation speed increases from 2.15 FPS (frames per second) to 5.88 FPS. Meanwhile, the segmentation accuracy (AP50) increased from 90.56% to 91.21%. In addition, it was proven that the particle size distribution and shape could be analyzed accurately and rapidly with the proposed model, providing essential information for multiphase flow process optimization and equipment design in industrial applications.

Abstract Image

查看原文本刊更多论文

用于多相流中实例分割和颗粒物理性质分析的轻量级掩模 R-CNN

本文开发了一种轻量级掩膜 R-CNN 实例分割模型，可准确快速地分析颗粒的大小和形状。首先，提出了混合深度稀释卷积网络（DDNet），并简化了特征金字塔层和区域提议网络的共享卷积层，在降低模型复杂度的同时保证了强大的特征提取能力。然后，通过引入 Dice 损失函数和聚类算法，在不牺牲计算速度和性能的前提下显著提高了分割精度。实验结果表明，模型参数大幅减少了 49.46%，分割速度从 2.15 FPS（帧/秒）提高到 5.88 FPS。同时，分割准确率（AP50）从 90.56% 提高到 91.21%。此外，研究还证明，利用所提出的模型可以准确、快速地分析粒度分布和粒形，为工业应用中的多相流工艺优化和设备设计提供重要信息。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Powder Technology 工程技术-工程：化工

CiteScore

9.90

自引率

15.40%

发文量

1047

审稿时长

46 days

期刊介绍： Powder Technology is an International Journal on the Science and Technology of Wet and Dry Particulate Systems. Powder Technology publishes papers on all aspects of the formation of particles and their characterisation and on the study of systems containing particulate solids. No limitation is imposed on the size of the particles, which may range from nanometre scale, as in pigments or aerosols, to that of mined or quarried materials. The following list of topics is not intended to be comprehensive, but rather to indicate typical subjects which fall within the scope of the journal's interests: Formation and synthesis of particles by precipitation and other methods. Modification of particles by agglomeration, coating, comminution and attrition. Characterisation of the size, shape, surface area, pore structure and strength of particles and agglomerates (including the origins and effects of inter particle forces). Packing, failure, flow and permeability of assemblies of particles. Particle-particle interactions and suspension rheology. Handling and processing operations such as slurry flow, fluidization, pneumatic conveying. Interactions between particles and their environment, including delivery of particulate products to the body. Applications of particle technology in production of pharmaceuticals, chemicals, foods, pigments, structural, and functional materials and in environmental and energy related matters. For materials-oriented contributions we are looking for articles revealing the effect of particle/powder characteristics (size, morphology and composition, in that order) on material performance or functionality and, ideally, comparison to any industrial standard.