A Method for Lichen Cover and Bark Thickness Estimation from Tree Log Images
Dogan, Burak (2025)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-202505059108
https://urn.fi/URN:NBN:fi:amk-202505059108
Tiivistelmä
The aim of the project was to develop a method to extract data from tree log images. This project focused on detecting lichen area cover and calculating bark thickness from images of tree logs. The extracted data will be used by Predictive Community Ecology Research Group in University of Jyväskylä.
This is the first step to a longer project. Data extracted from this stage will be further used in future steps. For cropping Segment Anything model (SAM) by Meta was used. Then images were normalized using HSV background normalization method. For estimating lichen areas, pixel HSV data of lichens were collected through GIMP and used for detecting areas in the given range. Finally, a bark thickness estimation method was developed using a derivative approach by fitting polynomials to observed bark thickness.
Results indicate methods are working. Images were successfully cropped using SAM algorithm. For failed cases, use of different model checkpoints was suggested. Normalization method worked as expected, improving lighting conditions. Lichen area estimation by using HSV ranges of lichen pixels working as expected. For extreme cases such as different lighting conditions, solutions were suggested to improve accuracy of the estimation. Bark estimation method testing revealed that both HSV saturation component and value component need to be used, and estimations need to be confirmed by analysing the plots. Some cases contrast adjustment improves accuracy but needs to be applied in cases of wrong estimation.
Overall, methods developed for each step worked as expected. Data extracted through this method can be used for further studies. Suggestion for improving accuracy, and workarounds for possible issues were provided.
This is the first step to a longer project. Data extracted from this stage will be further used in future steps. For cropping Segment Anything model (SAM) by Meta was used. Then images were normalized using HSV background normalization method. For estimating lichen areas, pixel HSV data of lichens were collected through GIMP and used for detecting areas in the given range. Finally, a bark thickness estimation method was developed using a derivative approach by fitting polynomials to observed bark thickness.
Results indicate methods are working. Images were successfully cropped using SAM algorithm. For failed cases, use of different model checkpoints was suggested. Normalization method worked as expected, improving lighting conditions. Lichen area estimation by using HSV ranges of lichen pixels working as expected. For extreme cases such as different lighting conditions, solutions were suggested to improve accuracy of the estimation. Bark estimation method testing revealed that both HSV saturation component and value component need to be used, and estimations need to be confirmed by analysing the plots. Some cases contrast adjustment improves accuracy but needs to be applied in cases of wrong estimation.
Overall, methods developed for each step worked as expected. Data extracted through this method can be used for further studies. Suggestion for improving accuracy, and workarounds for possible issues were provided.