A novel feature extraction technique to retrieve vegetation class for fire risk assessment

This paper presents the work done in an attempt to develop an automatic computer vision system for the identification of dense and sparse grasses from the roadside. One of the major problems for the identification of this kind of grasses is similar spectral signature with respect to the color, shape and also irregular distribution in the field. However concerns regarding the application have prompted increasing interests in seeking a novel feature extraction technique that can be used in identifying dense and sparse regions from roadside. This paper describes a new strategy involving two processes: extracting features with novel feature extraction technique and decision making. The strategy consists of a novel feature extraction technique that is effective for extraction of features from the digital image while decision making is based on nonlinear Support Vector Machine (SVM) with a radial basis function. Analysis of the results reveals that proposed feature extraction technique with SVM achieves above 93% accuracy on test images.