Banana Fusarium Wilt disease detection based on UAV remote sensing.

Abstract

There is a significant threat to the global supply of bananas that is posed by the disease known as Panama wilt, which is also referred to as Banana Fusarium Wilt (BFW) in some instances. A significant dissatisfaction is caused by the fact that there are currently no viable treatment options available for BFW. The early monitoring of the disease and the evaluation of its distribution were the primary focuses of this research project, with intention of making a contribution to the decrease of BFW. A model that is capable of identifying regions or plants within a banana plantation that are either infested without BFW or free of BFW was developed as one of the specific objectives. The other specific objectives include the identification of an appropriate image processing technique, the determination of sensitive parameters for the selected technique, and the development of a model. Utilizing an unmanned aerial vehicle at a flying height of 20 m above the ground, multispectral images were captured over a BFW-affected banana plantation. A single flight, covering 3 acres, yielded images totaling 639 under standard operational conditions. The categorization model included two types of spectral features as inputs: three multispectral band images and one vegetative index (VI). A self-organizing data analysis approach was utilized to identify canopies that are infected with BFW. Comparative analysis demonstrated that canopies infected with BFW exhibited higher reflectance in the Normalized Difference Vegetation Index (NDVI) range and exhibited distinctive color variations in the NDVI region compared to canopies that were healthy. The research results indicated that VIs, such as NDVI were successful in accurately detecting BFW Disease. The study employed binary logistic regression to evaluate the spatial correlation between VIs and the presence or absence of BFW in plants. The algorithm effectively detected the disease and accurately delineated specific regions using landmarks. The study further employed Google Maps to quantify the distances between afflicted plants and nearby landmarks. The research findings provide valuable information for management of banana plantations, presenting practical methods for detecting plant diseases and providing recommendations to farmers in reducing the impact of Panama wilt disease on banana cultivation.