FAST CLASSIFICATION OF LEAF IMAGES FOR AGRICULTURAL REMOTE SENSING APPLICATIONS

Abstract

This paper introduces a method of classifying leaves using machine learning. Considerable emphasis has been put on leaf classification for use in remote sensing applications such as plant phenotyping and precision agriculture. Convolutional neural networks (CNN) have been extensively used in computer vision for image classification. However, CNN can be computationally expensive. This paper describes a method that achieves a comparable accuracy, with a lower computational burden, using a support vector machine (SVM) classifier. This method uses image processing algorithms to extract features from Hough transform and Hough Lines. These features are then integrated with those extracted from binary images, and “eigenleaves” extracted from grayscale, gradient, and different color-space images of leaves as data preprocessing for classification. The classifier is implemented on two publicly available datasets: Flavia and Swedish; and is able to achieve state-of-the-art accuracies using a SVM classifier.