Automated deep learning approaches for classifying whitefly pest populations in cotton leaves during scouting operations

Monday, November 11, 2024

10:30 AM – 10:42 AM MST

Location: Phoenix Convention Center, 130, PCC

Presenting Author(s)

DK

Denis Olgen Kiobia

University of Georgia
TIFTON, Georgia

Co-Author(s)

CM

Canicius Joseph Mwita

University of Georgia
TIFTON, Georgia
Jason M. Schmidt

Associate Professor
University of Georgia
Tifton, Georgia
GL

Guoyu Lu

University of Georgia
Athens, Georgia
HS

Haijian Sun

University of Georgia
Athens, Georgia
MK

Mehmet Kurum

University of Georgia
Athens, Georgia
GR

Glen Rains

University of Georgia
Tifton, Georgia

Scouting of whitefly pests across various crop systems traditionally relies on identifying hotspot areas with high concentrations of adult and nymph whitefly populations. Unfortunately, most farmers estimate such populations based on human observation for management and control decisions. The approach of manual observation during scouting has several problems, including low accuracy, being tedious when dealing with densely populated plants, and difficulty in retrieving the scouting records. Recently, deep learning algorithms tailored for object detection have been proposed to alleviate this burden by estimating the whitefly population based on images. However, studies evaluating the whitefly population in multiple cotton leaves under field conditions are limited. In addition, the information on whether a deep learning algorithm detects young and whiteflies in cotton fields is narrow. This study investigated the feasibility of getting a deep-learning architecture to estimate whiteflies during scouting operations in a cotton environment. Particularly, the efficacy of different automated classification deep learning algorithms, including Convolutional Neural Network, ResNet, YOLOv4, and Faster R-CNN were evaluated. Our experimental results indicate that Deep learning algorithms can detect young and adult whiteflies in multiple leaves.