Comparison of Support Vector Machine and k-Nearest Neighbor Classifiers for Tree Species Identification
Tree species are a group of organisms that have different characteristics, and the identification of tree species has become essential to protect biodiversity, ecosystem balancing, as well as for medicinal purposes. It takes too much time for experts to identify individual tree species, so research based on automatic tree species identification is required to save time. Remote sensing-based satellite Image Processing (IP) techniques are useful for detecting features of different tree species like shape, texture, color, etc., but IP techniques alone are not sufficient for automatic detection of tree species. So, some Machine Learning (ML) algorithms are useful for the identification of tree species. Here, Histogram of Oriented Gradient (HOG) features have been introduced to identify different tree species like, banana, coconut, lemon, mango, oil palm, papaya trees, and those which do not belong to any category, are recognized as a not specified class. So, the combination of IP and ML techniques are useful. Here, comparison of k-Nearest-Neighbor (kNN) and Support Vector Machine (SVM) classifiers with HOG has been introduced, and overall accuracy (OA) for kNN and SVM can be obtained 85.71% and 93.33% respectively. For further evaluation, K-fold cross-validation is used, and it gives 88.39% and 94.34% OA for kNN and SVM respectively.
HOG feature descriptor;K fold cross-validation;k-Nearest-Neighbor (kNN);Support vector machine (SVM);Tree species identification
Full Text: PDF (downloaded 270 times)
- There are currently no refbacks.