02079naa a2200361 a 450000100080000000500110000800800410001902400540006010000180011424500960013226000090022852010630023765000260130065000130132665000290133965000290136865000170139765000160141465000090143065300250143965300180146465300250148265300210150765300090152870000160153770000160155370000200156970000180158970000240160770000160163170000220164777300480166921660072024-07-29       2024    bl uuuu     u00u1 u    #d7 ahttps://doi.org/10.1016/j.cropro.2024.1068462DOI1 aTETILA, E. C.  aReal-time detection of weeds by species in soybean using UAV images.h[electronic resource]  c2024  aABSTRACT. In this work, we evaluated a family of You Only Look Once (YOLOv5) object detection models for real-time detection of weeds in soybean fields. Based on the results generated by the detection, agricultural inputs can be applied to the identified regions of interest, lowering production costs by reducing the average number of sprayings and contributing to ecological balance and environmental preservation. We used the UAV to fly over three agricultural areas at an altitude of 10 m. Thus, we created a new dataset of 4129 annotated plant samples, which can serve as a baseline for weed detection in soybean crops. We considered four metrics to evaluate the classification results and three to evaluate detection results. Experimental results showed low average error rates in almost all test scenarios. YOLOv5s6 produced the best results among the evaluated models, obtaining MAE, RMSE, and R2 rates of 1.14, 1.67, and 0.93, respectively. We also demonstrate how the model can be deployed as part of an end-to-end system for herbicide application.  aPrecision agriculture  aSoybeans  aUnmanned aerial vehicles  aAgricultura de Precisão  aErva Daninha  aGlycine Max  aSoja  aAprendizado profundo  aDeep learning  aDetecção de objeto  aObject detection  aWeed1 aMORO, B. L.1 aASTOLFI, G.1 aCOSTA, A. B. da1 aAMORIM, W. P.1 aBELETE, N. A. de S.1 aPISTORI, H.1 aBARBEDO, J. G. A.  tCrop Protectiongv. 184, 106846, Oct. 2024.