Digital image processing based identification of nodes and internodes of chopped biomass stems

• Chopped biomass stem nodes and internodes can be identified from gray scale images.
• Narrow computational band along the major axis was sufficient for identification.
• Nodes have greater variation in gray value along major axis than internodes.
• Unit area under the normalized gray value curve gave best identification.

Chemical composition of biomass feedstock is an important parameter for optimizing the yield and economics of various bioconversion pathways. Although chemical composition of biomass varies among species, varieties, and plant components, there is distinct variation even among stem components, such as nodes and internodes. Separation of morphological components possessing different quality attributes and utilizing them in ‘segregated processing’ leads to better handling, more efficient processing, and high-valued products generation. Using equipment to separate morphological components such as node and internodes of biomass stem that have closely related physical properties (e.g., size, shape, density) is difficult. However, as the nodes and internodes are clearly distinct in appearance by visual observation, the potential of digital image analysis for node and internode identification and quantification was investigated. We used chopped stems of big bluestem, corn, and switchgrass as test materials. Pixel color variation along the length was used as the principle of identifying the nodes and internodes. An algorithm in MATLAB was developed to evaluate the gray value intensity within a narrow computational band along the major axis of nodes and internodes. Several extracted image features, such as minimum, maximum, average, standard deviation, and variation of the computational band gray values; ribbon length of the computational band normalized gray value curve (NGVC), unit ribbon length of NGVC; area under NGVC, and unit area under NGVC were tested for the identification. Unit area under NGVC was the best feature/parameter for the identification of the nodes and internodes with an accuracy of about 96.6% (9 incorrect out of 263 objects). This image processing methodology of nodes and internodes identification can form the supporting software for the hardware systems that perform the separation.