Object Detection in Differing Seasonal Conditions : Detection of leaves on a tree
Pöllänen, Samu (2025)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202502032212
https://urn.fi/URN:NBN:fi:amk-202502032212
Tiivistelmä
The purpose of this thesis was to create a model that could segment the leaves of European Oak trees on images. The technical approach was to use synthetic data created with SideFX Houdini, together with real-world images to expand the dataset. The architecture chosen was YOLO v8 for its ease of use.
In previous studies conducted on the subject, most of the studies used only realworld images. The aim of this study was to prove that the use of synthetic data could create a model with a smaller real-world dataset that still performs well.
The model that used a combined data set outperformed both models that used only one type of data in this thesis. However, the data sets remained rather small and the variation between images was quite low, weakening the generalisability of the model. In future studies using a larger and more high-quality real-world dataset could lead to even better performance of the model.
In previous studies conducted on the subject, most of the studies used only realworld images. The aim of this study was to prove that the use of synthetic data could create a model with a smaller real-world dataset that still performs well.
The model that used a combined data set outperformed both models that used only one type of data in this thesis. However, the data sets remained rather small and the variation between images was quite low, weakening the generalisability of the model. In future studies using a larger and more high-quality real-world dataset could lead to even better performance of the model.