An Optimal Wavelet Type to Increase Classification Accuracy for Forest Softwood Species

Abstract

Computer vision systems are a very interesting alternative approach for the forest species classification, which is a challenging task that generally requires well-trained human specialists. Hence this work proposed a method concerned with computer vision system to classify 37 different forest softwood species. The images are being decomposed to red, green and blue channels to extract features, each channel is decomposed using a different discrete wavelet transform (DWT) family. The image features are prepared to be classified. Texture images of softwood species are classified using neural network (NN) and support vector machine (SVM) classifiers. Classification accuracy (CA) values for SVM were 94.7297% and 97.2973% for 3rd and 5th level decompositions “coif2” DWT respectively which are better than NN results.