Métodos convencional e de aprendizagem de máquina na identificação de madeira como instrumentos para o controle do comércio de espécies tropicais

Abstract

Deforestation and illegal timber trade are major threats to the planet's sustainability. International efforts to prevent the spread of these environmental crimes include the creation of laws, commercial restrictions for certain species and inspection. However, such measures have low effectiveness in practice, laws and official documents can be defrauded, since several threatened woods are freely traded in commerce. The objective was to evaluate the accuracy of traditional identification methodologies and technological approaches in classifying ten timber species available at Brazilian retail, including species threatened with extinction. It was also determined which anatomical patterns demonstrate the correct identification of eight wood species from the amazon region. The results obtained in the first work showed the potential of classification based on Convolutional Neural Networks (CNN), which achieved global accuracy greater than 95% in identifying threatened native wood Amburana cearensis (100%), Bertholletia excelsa (99%), Cedrela odorata (96%). This model based on the Inception V3 algorithm was also able to accurately classify taxa of high commercial value Aspidosperma sp. (100%), Diplotropis sp. (98%), Dypterix odorata (99%) and Hymenolobium petraum(100%). Despite this, the model had difficulty identifying similar taxa Erisma uncinatum (95%), Guareasp. (94%) and Vochysiasp. (82%), wood commonly grouped by the name “cedrinho” in the region of Lavras, Minas Gerais state. From the macroscopic analysis of similar woods, it was observed that the axial parenchyma was the determining pattern to separate the studied taxa. Apparently, the wood anatomy structures function in patterns, which we used to determine their botanical identify. Based on anatomical information and other relevant data on the species, in this thesis we present Smart Timber ID, a digital wood identity that seeks to provide transparency and evidence for the traceability of origin the forests products. It is expected that the integration of technological and traditional methods, as well as the availability of this digital document, will promote the sustainability of the international tropical timber market, thus helping the conservation of several native species and promoting awareness consumption.