Fully automated method to estimate opacity in stack and fugitive emissions: A case study in industrial environments

Abstract

Fugitive emissions are those that are unplanned, i.e., they do not come out of a stack. These emissions are usually disperse and difficult to locate. By estimating the opacity of fugitive emissions they can be controlled or even prevented, helping to comply with environmental regulations. Most opacity estimation methods are based on stack emissions, which are straightforward, as they are always located in the same area. All methods in the literature for emission opacity estimation require a human operator to select the regions to be used as a reference. In this work, deep learning networks are proposed to improve the quality and automation of this process by selecting the regions completely and automatically. Furthermore, a new fugitive emission opacity estimation method is proposed. This method, called SBPB, is compared with other relevant methods in the literature, offering a solution with an average F1-Score metric 5 % higher than other methods on two real datasets with over 4000 images in total. This method provides a robust solution for fugitive emissions.