The data consists of data from the Berry Machine project. The project has aimed to develop an AI-based system to improve berry observations and forecasts in observation forests. The Natural Resources Institute Finland has been making observations and forecasts of berry harvests since the 1990s. The observations have been based on plots in the forests. Each forest has five observation plots of one square metre. Flowers, raw berries and ripe berries are counted during the growing season and forecasts are based on these calculations. The aim of the Berry Machine project was to develop a machine-learning system that can be used to estimate the quantity and quality of berry harvests using a photo taken with a smartphone. The target was blueberry (Vaccinium Myrtillus) and lingonberry (Vaccinium Vitis-idea) A prototype application was implemented on a smart phone to determine the number of bilberry and lingonberry flowers, raw berries and ripe berries using machine vision. To determine the area of the observation area needed for berry density, AR-based measurement was tested in the prototype. The ratio of the number of berries interpreted from the image to the actual value was compared with the help of the field measurements. The number of berries detected by machine learning and the result of the field measurement were attached to the same image. For the study of berry density estimation using an image, the area of berries detected by machine learning was determined by cropping the detected berries by the boundaries of the (annotated) count frame drawn on the same image. The bounding with polygon verifies that berries have been detected in the image are exactly in the area corresponding to the area of the frame. The dataset is in three parts: 1. 640x640 Images with VOC annotations. 640x640 annotated images ready for machine learning. 2. datatables: datasheets containing metadata about the images, inference results of the trained machine learning model on the images, and information about the count frames and images with field measurement berry density 3. Original. Original image files and annotations (in COCO format). 80GB of original material is split into eight zip files for easy downloading. Also frame annotation polygons (COCO-format) The data in the tables can be linked to each other via the image file path. For a more detailed description of the data, see README.md. The project's AI system is prototyped and implemented by the FrostBit Software Laboratory at Lapland UAS.

2022