Identification of wood from the Amazon by characteristics of Haralick and Neural Network: image segmentation and polishing of the surface

• Published in: IForest (Viterbo) Vol. 15; no. 4; pp. 234 - 239
• Main Authors: de Souza Vieira, GL, Moutinho da Ponte, MJ, Pereira Moutinho, VH, Jardim-Gonçalves, R, Pantoja Lima, C, de Albuquerque Vinagre, MV
• Format: Journal Article
• Language: English
• Published: Potenza The Italian Society of Silviculture and Forest Ecology (SISEF) 01.08.2022; Italian Society of Silviculture and Forest Ecology (SISEF)
• Subjects: Algorithms; Amazon; Artificial Neural Networks; Back propagation; Biodiversity; Classification; Decision making; Digital Image Processing; Finite element method; Forests; Identification; Identification methods; Image processing; Image segmentation; Inspection; Neural networks; Object recognition; Pattern Recognition; Polishing; Rainforests; Regression models; Sandpaper; Species; Technology; Training; Wood; Wood Identification; Brazil; Amazon Basin
• DOI: 10.3832/ifor3906-015
• ISSN: 1971-7458

The identification of Amazonian timber species is a complex problem due to their great diversity and the lack of leaf material in the post-harvest inspection often hampers a correct recognition of the wood species. In this context, we developed a pattern recognition system of wood images to identify commonly traded species, with the aim of increasing the accuracy and efficiency of current identification methods. We used ten different species with three polishing treatments and twenty images for each wood species. As for the image recognition system, the textural segmentation associated with Haralick characteristics and classified by Artificial Neural Networks was used. We verified that the improvement of sandpaper granulometry increased the accuracy of species recognition. The developed model based on linear regression achieved a recognition rate of 94% in the training phase, and a post-training recognition rate of 65% for wood treated with 120-grit sandpaper mesh. We concluded that the wood pattern recognition model presented has the potential to correctly identify the wood species studied.