ERWIAM dataset

Description

Annotated image dataset of Fire blight symptoms for object detection in orchards. Monitoring plant diseases in tree nurseries, breeding farms and orchards is essential for maintaining plant health. One of the most dangerous diseases in fruit growing is fire blight (Erwinia amylovora), as it can spread epidemically and cause enormous economic damage. In this context, a digital disease monitoring system for fire blight based on RGB images was developed for orchards. Data was collected on a total of nine dates in 2021, 2022 and 2023 under different weather conditions and with different cameras. The RGB images were taken of different apple genotypes after artificial inoculation with Erwinia amylovora, including varieties, wild species and progeny from breeding. The presented ERWIAM dataset contains manually labelled RGB images with a size of 1280 x 1280 pixels of fire blight infected shoots, flowers and leaves at different stages of development as well as background images without symptoms. In addition, symptoms of other plant diseases were recorded and integrated into the ERWIAM dataset as a separate class. The dataset contains a total of 1,611 annotated images and 87 background images. This dataset can be used as a resource for researchers and developers working on digital plant disease monitoring systems.

Steps to reproduce

The RGB raw data was collected on six days between 2021 and 2023 at different data source locations. Data collection took place in the experimental orchard and in the experimental greenhouse. The data source location for the symptoms of Erwinia amylovora infection was the experimental orchard of the Julius Kühn Institute (JKI) at the Institute of Plant Protection in Fruit Crops and Viticulture in Dossenheim, Germany, and the experimental greenhouse of the JKI for Resistance Research and Stress Tolerance in Quedlinburg, Germany. In addition, the raw RGB background images were collected on three days in 2022 and 2023 in the JKI's experimental orchard for breeding research on fruit trees in Dresden-Pillnitz. The data was collected using various cameras, including smartphones such as Samsung Galaxy S10 version SM-G973F (2019), Samsung Galaxy S20 FE 5G version SM-G781B (2020) and Samsung Galaxy Tab A 10.5 tablet version SM-T590 (2019), all from Samsung Electronics Co., Ltd, as well as the Canon EOS 70D version 1.1.1 (2013) and Canon EOS 90D version 1.1.1 (2019) from Canon Incorporated. The image metadata was saved in an Excel spreadsheet using P. Harvey's ExifTool software version 12.3.8.0 (2016). After data collection, the images were sorted manually, taking into account the image quality and the occurrence of fire blight symptoms on flowers, leaves and shoots. All infection stages were categorised as "FLOWER" (= blossom blight), "LEAF__" (= leaf blight) and "SHOOT_" (= shoot blight). Overlapping and nested fire blight symptoms were also labelled. In addition, symptoms of other plant conditions and potential fire blight symptoms that could not be clearly assigned were recorded and integrated into the ERWIAM dataset as a separate class "MAYBE_". For annotation, the CVAT tool version 1.1.0 by B. Sekachev et al. (2020) was used to perform 2-point bounding box annotations. After the annotations were completed, a labelling file (.txt) in YOLO 1.1 format was created for each image. A total of 15,761 annotations were manually created on 1,611 images. The image sizes were reduced to 1280 × 1280 pixels in pre-processing using the Python library Pillow version 9.5.0 (Lanczos filter) by J. A. Clark (2015) and saved in JPG format. In addition, 87 background images without disease symptoms were added. The use of different camera settings, varying distances to the objects and different shooting angles contributed to the diversity of the dataset. Fire blight symptoms were also annotated if they showed light reflections, shading or moisture in addition to the fire blight infection.

Institutions

Categories

Funding

Licence