Spectral mixture analysis based citrus greening disease detection using satellite image of florida

Abstract

The citrus industry in Florida is being threatened by some devastating diseases, one of which is called citrus greening, also known as Huanglongbing (HLB). Timely and location-specific detection and monitoring of the infected citrus trees are required for efficient disease control, while reducing risks of the disease being spread. Satellite images were used to detect citrus greening disease over large areas in this paper, because it has larger view scale and less costly compared to airborne hyperspectral (HS) and multispectral (MS) images. Firstly, normalized difference vegetation index (NDVI) was used to remove water and bare soil from the image. Then, support vector machine (SVM) was used to separate citrus groves from other natural vegetation. Two different endmember extraction methods, vertex component analysis (VCA) and iterated constrained endmember detection (ICE) were used to extract endmembers for soil, grass, HLB infected citrus trees, and healthy citrus trees from endmember candidates identified from a Landsat 5 Thematic Mapper (TM) image with the help of a WorldView-2 image. These endmembers were then used as an input of multiple endmember spectral mixture analysis (MESMA), in order to obtain the fraction map of the citrus trees. The results were compared and evaluated. It was demonstrated that there is a great potential for citrus greening disease detection using a satellite image.