外文摘要：Earlier, researchers have employed computer vision-based automatic fruit detection to estimate coffee yield at the time of harvest, however, studies on the on -plant quantification of the coffee fruit are scarce. In this study, the latest version of the state -of -the -art algorithm YOLOv7 (You Only Look Once) was used for the first time. YOLOv7 was trained with 324 annotated images of fruit bearing coffee branches followed by its evaluation with 82 annotated images as validation data (supervised method) and then tested through raw images (unannotated) as test data. Meanwhile, the K-means models were trained which led to machine-generated color classes of coffee fruit for semi-supervised image annotation. Consequently, the developed model efficiently analyzed the test data with an mAP@.5 (mean average precision) of 0.89. Strikingly, our innovative semi-supervised method with an mAP@.5 of 0.77 for multi-class mode surpassed the supervised method which had mAP@.5 of only 0.60, leading to faster and more accurate annotation. While testing the yield estimation of the model in two plots, an average error of 3.78% was recorded between the predicted (pre-harvest) and ground truth (harvest) data for binary class mode. The average error for multi-class data was recorded as 3.87% for green, 3.445% for green-yellow, 5.09% for cherry-raisin, and 2.51% for dry fruits. This AI-based technology when integrated with other tools such as UAV would efficiently remotely monitor coffee field for informed decision about irrigation, fertilizer application and other measures of timely field management, hence advancing precision agriculture. Moreover, this machine learning intelligent model can be tailored for various other fruit farming.
外文关键词：precision agriculture;machine learning;artificial intelligence;coffee;digital phenotyping
作者：Eron, Francisco;Noman, Muhammad;de Oliveira, Raphael Ricon;Chalfun-Junior, Antonio
作者单位：Univ Fed Lavras
期刊名称：SCIENTIA HORTICULTURAE
期刊影响因子：0.0
出版年份：2024
出版刊次：327
原文传递申请：江苏省科技资源（工程技术文献）统筹服务平台