Moisture content recognition for wood chips in pile using supervised classification

Abstract

Wood chips moisture content (MC) is a key parameter for controlling the biofuel product qualities and properties. Since no knowledge-based model is available to recognize MC, machine learning methods are promising techniques to design black-box models for MC prediction or recognition. As wood permittivity strongly changes in the presence of water, an electromagnetic module is used to probe the reflectivity of wood chip piles. In the present paper, the recognition of three wood chip piles of different MC categories is performed using support vector machines (SVMs). SVM-recursive feature elimination is implemented to rank and select reflection coefficients to design optimized linear SVM classifiers that attribute MC class of wood chips in a pile. Experiments show that the proposed approach is effective and requires a limited computational power. The global classification performance exceeds 95%.