Cotton Root Rot Identification & Delineation Based on UAV Platform
Abstract
Cotton Root Rot (CRR) is a severe cotton disease in Texas and the southwestern U.S. It kills cotton and other dicotyledonous plants so quickly that the death of the plant may be the first observable symptom. CRR cannot be cured within season, but spraying flutriafol fungicide next to the seed during planting controls the disease effectively. A previous study indicated that CRR typically reoccurs at the same location in fields, so historical location information can guide the grower on where to apply the fungicide more efficiently and to minimize environmental risk. Manned aircraft remote sensing has been proven capable of providing CRR location distribution information.
Because image data from manned aircraft tend to be expensive and have low spatial resolution compared to images from unmanned aerial vehicles (UAVs), three studies were undertaken to develop the capability of UAV remote sensing to determine CRR location in cotton fields so that fungicide can be applied in a precise manner. Study I was conducted to explore the possibility of using the UAV based remote sensing data to delineate CRR-infested areas and generate a prescription map. The result demonstrated that UAV remote sensing can be used effectively for this purpose and can significantly reduce the amount of fungicide applied.
Study II was conducted to develop methods to improve the image processing methods used to identify CRR in UAV remote sensing data in an effort to take advantage of their inherently high spatial resolution. Conventional classification methods used with manned-aircraft data do not work well on UAV data, because the higher resolution results in additional classes of image pixels like bare soil and shadows between crop rows. In this study, two new regional classification methods were developed which can accurately and automatically identify CRR-infested areas with high-resolution UAV remote sensing data. The results demonstrated that the new proposed methods are superior for CRR detection in UAV images compared to conventional classification methods.
Study III was an attempt to make further use of the high resolution of UAV data by creating a plant-by-plant level CRR identification method. The desire is to make fungicide application as precise as possible, even potentially at the level of individual plants, so as to minimize the amount applied, saving cost and reducing environmental risk. The results of this study illustrated that the plant-by-plant image classification method can identify individual plants and determine whether they are infected or not with high accuracy.
All of these studies were funded by Cotton Incorporated, Cooperative Research Agreement number 16-233.
Subject
precision agricultureUAV
disease detection
cotton root rot
plant-level
single-plant
plant-by-plant
classification
image analysis
machine learning
prescription map
semi-supervised
Citation
Wang, Tianyi (2020). Cotton Root Rot Identification & Delineation Based on UAV Platform. Doctoral dissertation, Texas A&M University. Available electronically from https : / /hdl .handle .net /1969 .1 /192372.