Sensoriamento remoto na identificação e caracterização de culturas de interesse zootécnico

Abstract

The objective of this study is to provide an integrative review on the use of remote sensing in the identification and characterization of crops of zootechnical interest; and evaluate the accuracy of the YOLO-v8 model in detecting corn plants. The first chapter aimed to identify the main technologies used in differentiation for agronomic characterization of crops. The review protocol was developed in which the search terms were previously tested and based on the PVO strategy, where population (P): corn (Zea mays), millet (Pennisetum sp.) and sorghum (Sorghum bicolor) crops; the variables (V) of interest: were technologies used to obtain images and methodologies applied to processing; results (O): agronomic variables and differentiation). The guiding question was: What are the main technologies used to differentiate corn, sorghum and millet crops for agronomic characterization? The databases used were Scopus (Elsevier), Web of Science (Main Collection), Science Direct and CAB direct and the search was carried out until September 14, 2022. The search identified 585 research articles; however, only 35 were considered eligible and included due to their methodological robustness for data extraction. The most used crop was corn (80.00%). Among the studies, the remote sensing platforms most used to obtain images were those based on UAVs (74.28%). The most used image processing software was Agisoft Metashape (37.14%). Agronomic variables (82.85%) were most addressed, namely biomass, height and productivity. In the second chapter, the objective was to evaluate the YOLO-v8 model in detecting corn plants in the initial growth stage in order to enhance crop production. A data set was constructed to evaluate the selected method, a proportion of 60% of the data was used for training (338 image clippings), 20% for testing (113 image clippings) and 20% for validation (113 image clippings), totaling 564 image clippings and 39765 labels. The mAP0.5 value found in this study was 90.5%, indicating that the YOLO-v8 method constitutes a promising approach for developing tools for estimating corn plants and agricultural monitoring.