Feature Selection for Plant Leaf Classification Based on Information Gain

Feature selection methods have been explored in the literature for the classification techniques, among which correlated feature, information gain, mutual information and chi-square are considered more effective. The leaf images contain inherent noise due to imaging equipment, operating environment and position of the image during image acquisition. In this paper, a method for classification of leaf images is proposed by exploiting the concept of information gain and explores the efficacy of learning algorithms of Multi-Layer Perceptron (MLP) for classifying plant leaf.This research shows information gain method for MLP with Batch Back propagation algorithm based learning increases computational efficiency by improving classification accuracy.It is observed that the proposed measure outperforms MLP with incremental training and Levenberg Marquardt based learning for plant leaf classification when tested with 9 species. Evaluation illustrates that information gain helps select features that result in significant improvements on MLP with Batch Back propagation algorithm classifier performance with an accuracy of 94.81%. Keywords—Feature selection, Plant Leaf Classification, Information Gain, Multi-Layer Perceptron (MLP)