Identification of Biomass Powder Fuel Based on Image Processing and Feature Engineering

Abstract

Biomass fuels have a wide application prospect in thermal power generation. Different fuels have different physical and chemical properties and are reflected in the complex combustion and pollution emission properties during application. It is necessary to determine the type of biomass fuel. This paper proposes an image-based method to identify the type of biomass fuel and some related work derived from it. A feature selection rule that can explain the physical meaning of the fuel was designed. Through an ensemble learning method named Stacking and an image partition recognition strategy, the recognition task is integrated from the perspective of algorithms and data. The 10-fold cross-check on the training set has an accuracy of more than 98%, and the effectiveness of the image partition recognition strategy was proved mathematically.