Limnia Pilot Kavala Dataset Repositories from Quadruped and Drone Version2

This is the v2 of the dataset reagrding Drone tools HD and multispectral used in Kavala pilot, which includes: 1) the YOLO repository models used for inference of tree crowns (Filename Drone_Crown_YOLO_Infr_models.7z) and a dataset set for evaluation was used Drone_Crown_YOLO_Infr_models.7z 2) Crown images for diseased trees that were found at the SeihSou forest nearby Thessaloniki for testing validation can be found at (Filename Deseased_trees_Testing_Validation_Forest_SeihSou.zip), this is because in the Limnia forest polygon, we found only 2 trees completely dead trees but we did not found any heavily diseased trees, either with brownish foliage or with depleted foliage, and we were forced to search in nearby other forests. The images were used to test and validate YOLO models' Tdead class representing preclassified diseased trees. (with preclassification we mean that the tree is under a disease or heavy stress). Worth mentioning here that as the drone flies at low altitude, the contribution of multispectral vs the simple RGB images is limited as the resolution of the RGB is HD quite high. In other words, HD RGB images for YOLO disease detection (low foliage or none or brownish foliage as a general preclassification grouping) are more than enough to achieve high detection metrics. 3) Regarding datasets with multispectral images for testing & validating water bodies and water streams, as such were not present inside the Limnia Forest pilot we search in other forests and specifically found water bodies at the forests of Thermi Dam, at the dams Thisauros, Purnani, Illarion [WaterMultiSpectral_Purnari_Thisauros.zip, Water_mutlispectral_Fragma_Thermis.zip, ThermiForest_WaterBodies.zip]. Regarding water streams, we found an artificial stream in Forest Thermi Dam and the corresponding file is ThermiForest_WaterStream.zip. More information about water bodies testing and accuracy can be found at https://doi.org/10.5281/zenodo.19695127 4) Regarding water multispectral results of the pilot at Limnia forest Kavala the pilot did not had any water bodies or waterstreams. Therefore, for testing validation other datasets were used as mentioned above. Still, for the completeness of the post the zipped file LimniaWaterIndexGroup1_repo.7z is included using a water multispectral combo index criteria of AWEI and NDVI . Also, repositories indicating tiles with tree-healthy are included LimniaHealthTreesGroup1_repo.7z . Regarding the preclassification of diseased trees with multispectral, the pipeline uses a combination of vegetation indices the GNDVI and NDRE. The threshold can be set strictly or not in this case below GNDVI<0.1 (low chlorophyll) and NDRE <0.1. Worth mentioning that multispectral display the spectral absorbance on the tree's foliage, we are not talking here about YOLO, so the preclassification based on the mulispectral threshold methodology must be performed over various continuous seasons as the indication may simply present not enough water or nutrition. The tiles-areas that persist over time in stress indication are areas on disease with a degree of accuracy above 90%, and of course, these resulting areas finally have to be compared with neighboring areas to avoid false alarm over large time periods with low levels of rain. A repository with such image tile-areas results from the Limnia pilot over a specific time are stored at file StressTree_Limnia_MultiSpectral.zip. Finally a repository with the bare soil areas of the pilot where the quadruped can operate, found by using multispectral indices is included LimniaBareSoilGroup1_repo.7z. In this way, in a simple and automated way, the trails' gps layout can be found for a quadruped to run its patrol in a forest. 5) A last annotated repository, an early spring collection of crowns at the pilot of Limnia can be seen at the file Limnia_Spring_Image_Annotation_Crowns_Equalized.7z, it also has some indicative annotated sample images and a text file with the labels and the folder with the YOLO format annotation.