Waterbird detection, classification, and mapping

This dataset contains all data needed to replicate the study done by Bogaert et al. in the paper called "Automated Waterbird Detection, Classification and Mapping from UAV Imagery Using a Two-Stage Deep Learning Pipeline". To replicate the study, please read the following instructions: Python 3.11.9 should be used to run the code files. All images to train, validate and test both the detection and classification models are included, but the labels are not. To be able to train the models, labelling still needs to be done beforehand. The scripts are compatible with Roboflow annotations in YOLOv8 format. Other labelling software can be used, but the scripts need to be adapted then. In the ZIP files "JULY.zip", "OCTOBER.zip", and "MARCH.zip". There are 24 folders called F1-F24, which include the full-sized original images used to validate the models and get the final projections on the orthomosaics. "YOLO_images.zip" contains both the train and validation tiles to train the yolo model; however, it does not include the labels. "Train_Valid_Test_images.zip" contains all full-sized images to train, validate, and test the models. The .csv files "manual_counts_XXXX" contain the manually counted birds from March, July, and October per flight. The .ipynb files are all the jupyter notebook files used to do the study. To train the YOLO model, "yolo_july_october_april_reworked" should be run. To train the classification model, "CNN_reworked" should be run. To get the results per month (detections, classifications, and projections), "YOLO_CNN_XXX_reworked" should be run per month. "YOLO_real_results_reworked" contains the code to get the performance curves of the YOLO model, as well as the additional analyses (size, entropy, and brightness analysis). "GSD_test" contains the code to get the GSD analysis. Images for this analysis can be requested by contacting the author, since they are too large for this dataset. "March_metrics" and "March_test" contain all the code for the unseen dataset test. "RESULTS" contains the code to get the prediction bar plots per month. "Orthomosaics_F19-F23.zip" contains the orthomosaics of flight 19 until flight 23. This way, you can use them in QGIS to plot the projections of those flights by dragging and dropping the orthomosaics with GeoJSON files into QGIS.