Data-driven application for analyzing apples in apple orchards

marek studzins / unsplash

The aim of the Appilyzer project is to design and develop a data-driven application that will enable automatic analysis in the form of yield estimation and health assessment of apples on apple orchards. The accuracy and reliability of the system will be evaluated to enable comparability with other approaches.

Our project is based on the research field of Agriculture 4.0, which deals with the automation of agricultural processes, among other things. The project is motivated by the fact that nowadays both the estimation of yields and the analysis of harvest quality are often still carried out manually. A part of the plantation is assessed and the result is then extrapolated to the rest of the plantation. This approach is prone to error due to the extrapolation and remains time-consuming and cost-intensive due to the high use of human resources.

To automate this process, high-resolution videos of all apple trees are first taken using a drone, which are then cut into a sequence of non-overlapping images to avoid double counting. The apples in each of these images are then cropped using a convolutional neural network. To estimate the yield, the cut apples are counted. To assess the quality of the harvest, another neural network is trained to classify the already cut apples into healthy and unhealthy apples. The health assessment is represented by the ratio of unhealthy to healthy apples. This summer, videos of the apple orchard at Nütschau Monastery were recorded several times with the help of a DJI drone.

And here the student group in their own words (German):