Image processing techniques for dust monitoring in industry:Controlled-environment experiments and feature-based concentration mapping
Shaofeng Wang
Jiangjiang Yin
Liwei Shi
Jiangyang Lei
Zilong Zhou
School of Resources and Safety Engineering,Central South University,Changsha 410083,Hunan,China;Hunan Key Laboratory of Resources Exploitation and Hazard Control for Deep Metal Mines,Changsha 410083,China
摘要：Dust pollution in industrial environments should be closely monitored for health and process control.Traditional mea-surement methods often require intrusive sensors and lack continuous real-time capability.A novel non-intrusive,image-based method was proposed for real-time dust concentration estimation.Using a controlled experimental setup with a high-concentration dust generator,images of dust-laden air across a wide concentration range(10-1000 mg/m3)were captured.For each image,the color and texture features were extracted as predictors of dust concentration.Then the poly-nomial regression was used to correlate these image-derived features with actual dust concentrations measured by standard instrumentation.The analysis revealed strong nonlinear relationships:the grayscale mean intensity correlates with dust concentration(R2=0.79),and selected texture features yield even higher correlation(R2＞0.82).These image-derived metrics effectively capture the scattering and attenuation of light caused by suspended dust,serving as reliable proxies for particulate mass.Combining multiple image features into a composite regression model further improved estimation accu-racy.These findings demonstrate that camera-based image analysis can reliably estimate dust concentration in real time,offering a low-cost,non-intrusive alternative to conventional sensors.The approach could facilitate automated calibration of dust-generation systems and enable continuous air quality monitoring in industrial settings.
机标关键词：processingenvironmentmappingimageconcentrationcontrolledexperimentsfeature-based
论文发表日期：2025-12-31
在线出版日期：2026-01-30（本平台首次上网日期，不代表文献的发表时间）
页数：17( 318-334 )
英文信息