Comparação entre o método MAXVER e Random Forest na identificação de Pinus sp. em áreas de campo úmido
In Brazil, Pinus sp. is a very common invasive plant and has an excellent adaptation to edaphoclimatic conditions in the Southeast and South regions of the country, especially in wet field areas. As an invasive plant, it can cause the exclusion of native species through competition, causing a loss of biodiversity and even changes in the structure of the ecosystem. Therefore, having tools that help in planning the mitigation of environmental impacts caused by these species is very important. Among these tools is Remote Sensing, which can identify and locate exotic and invasive plant individuals. Thus, the present work aimed to use very high resolution multispectral images to test and compare the MAXVER and Random Forest (RF) classifiers through a confusion matrix, more specifically, the Kappa Index, to understand which one has the best performance in the identification of isolated individuals of Pinus sp. in wet field areas. To do that, 5 classes were created, "Pinus" (Pinus sp.), "Nao_Pinus" (indicating all the remaining components of the landscape that do not fit into the other classes), "Agua" (indicating places with water, such as lakes, for example), “Solo” (indicating exposed soil) and “Sombra” (indicating shady locations), with which the two classifiers were performed using a 4-band image composition (NIR, Red Edge, Red and Green). Then, 500 random points were created, divided equally by each class, carrying the information of its classification performed by the classifiers and the one performed manually, this one taken as real. From this information, the Confusion Matrix was constructed, obtaining the Kappa Index. Both were classified as having moderate agreement, being MAXVER with an index of 0.5023 and RF with an index of 0.5264. Observing these values, noticed that RF showed greater agreement than MAXVER and, consequently, among these two classifiers, RF is suggested being better for the presented purpose, under the methodology used. It is noteworthy that, in addition to MAXVER and Random Forest, other classifiers can be used to identify individuals of Pinus sp. and other species, which may even show better performance for this purpose.
The following license files are associated with this item: