Texture analysis for the segmentation of sugar cane multispectral images

Abstract

In this paper is presented an analysis of the impact of texture features for segmentation of multispectral aerial images of sugar cane. Currently there are no precise techniques to estimate objectively areas of fallen cane and this causes significant losses in crop productivity and industrialization. For the real-ization of this work was made an image dataset. To build this dataset was im-plemented a software from which were obtained labeled regions in the images related to this agronomic phenomenon and then were extracted some texture features and a typical agronomic index (NDVI). The features related to segmen-tation task were analyzed with classical techniques such as Principal Compo-nent Analysis and Decision Trees. The results obtained show good performance to distinguish normal sugar cane versus fallen sugar cane but not between dif-ferent fallen sugar cane classes. However this approach was satisfactory to es-timate the normal and fallen sugar cane areas and this increase the information quality available to support agronomic decisions.