Naslov
Dataset of an apple orchard for object detection

Autori
Antolković, Ana Marija ; Skendrović Babojelić, Martina ; Vrtodušić, Rea ; Šatvar Vrbančić, Mihaela ; Petek, Marko ; Viduka, Antonio ; Karažija, Tomislav ; Fruk, Goran

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Digital Technologies in Agriculture, Book of Abstracts. No. 1/2022. / Lončarić, Zdenko ; Jović, Jurica - Osijek : Fakultet agrobiotehničkih znanosti Sveučilišta Josipa Jurja Strossmayera u Osijeku, 2022, 21-21

ISBN
978-953-8421-03-7

Skup
1st International Symposium on Digital Technologies in Agriculture (ISDTA 2022) ; 1st Satellite Workshop Digital Agriculture in Rural Area (DIGITAGRA 2022)

Mjesto i datum
Osijek, Hrvatska, 06.12.2022. - 08.12.2022

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
annotation, average precision, classes, computer vision, fruit

Sažetak
Computer vision has enabled the optimization of fruit production by providing an early alert to the farmer. Several automated crop monitoring systems have been proposed in the field of crop phenotyping, yield monitoring, and crop protection at various growth stages. Automated computer vision systems rely on large data sets to train, test, and compare diverse approaches. classification of different parts of apple tree fruit presents significant challenges due to interclass similarities because of the resemblance of some parts in the growth stages of apples. The objective of this research was to explore the possibility of detection of different apple tree parts during phenophases in an apple orchard using image analysis. The research was conducted in 2021 and 2022 in five commercial apple orchards around Zagreb in Croatia. About 5 000 images were taken with an RGB camera at different growth stages of the apple orchard. Images in the dataset were manually annotated in LabelImg software using bounding boxes and presented in PascalVOC format. Appels were annotated in different phenophases like a generative spur, flower cluster, healthy juvenile leaf, and various stages of the development of fruit. The EfficientDet architecture has been used as an object detection algorithm. Experimental results on a challenging dataset demonstrate that the average precision percentage of the generative spur, flower cluster, fruit frost, and deformed fruit was 51%, 63%, 38%, and 53%. Apple fruits were also classified by color, so we had classes for red and green apples whose average precision percentages for detection were 63% and 71%, respectively. The results indicate that various apple classes can be developed more effectively for detection, which can help the robot to decide the alert strategy (e.g., flower or fruit damage and yield planning) as well as to avoid potential damage by the branches and trellis wires.

Izvorni jezik
Engleski

Znanstvena područja
Računarstvo, Poljoprivreda (agronomija)

POVEZANOST RADA