Whitefly Detection in Agriculture Using Artificial Intelligence: A Systematic Review. (#653)

Authors

Nuñez Moran, Fabian Alejandro

Casaverde Pacherrez, Luis Alberto

Pingo Lozada, Jose Felix

Abstract

Agriculture is one of the most relevant economic activities globally, where pests such as the whitefly (Bemisia tabaci) generate significant crop losses. Traditional detection methods, in addition to being labor-intensive, present low effectiveness, which has driven the exploration of alternatives based on artificial intelligence (AI). The development of these technologies has enabled notable advancements in the early identification of B. tabaci, improving efficiency and reducing agricultural damage. However, their implementation faces operational challenges—such as foliage occlusion, lighting variability, and high equipment costs—as well as limitations in technical knowledge within the agricultural sector, particularly in Spanish-speaking regions. This systematic review (without meta-analysis) analyzes computer vision methods and drones for the detection and quantification of B. tabaci through a non-experimental descriptive design. Thirty open-access articles (2020–2025, in English, Spanish, or Chinese) were selected from seven databases (Scopus, SpringerLink, EBSCOhost, Web of Science, Scielo, ResearchGate, MDPI). These studies employed convolutional neural networks (CNNs), object detection models (Faster R-CNN, YOLOv4/v8), near-infrared spectroscopy (NIR), and hyperspectral imaging. The studies utilized CNNs, Faster R-CNN/YOLOv4/v8 detection models, NIR sensors, and various types of cameras, achieving >90% precision in the detection of adults, eggs, and pre-visual viral symptoms. The most effective techniques were CNNs (95.08%), YOLOv8 (87% on Raspberry Pi), and hyperspectral imaging (98%), consistently exceeding 85% precision. The optimal model selection depends on the type of analysis (leaf surface vs. internal symptoms), the pest's developmental stage, and crop characteristics, especially spectral variation. Hyperspectral systems excelled in early detection, while YOLO offered more accessible solutions for practical implementation.