An object dectection and classification model to identify Asian hornets, Vespa velutina nigrithorax, and European hornets, Vespa crabro, from images collected at bait stations. This builds the basis for an early warning system to track the spread of the invasive hornet Vespa velutina.
- explanation contains notebooks used for computing Layer-wise Relevance Propagation (LRP) heatmaps of trained model predictions.
- formatting contains notebooks & scripts used to generate YOLOv5-ready training/testing data from the raw Plainsight labelling outputs.
- images contains example graphics surrounding the project.
- manuals contains documentation detailing the Raspberry Pi implementation and usage.
- models contains notebooks for training YOLOv5s models
- yolov5-params contains trained weights over various datasets, as well as a bespoke augmentation file for the training scripts.
- monitor contains an implementation script monitor_run.py which imports YOLOv5 via PyTorch Hub and runs inference based on a webcam input. There are also various pre-filtering stages based on the ViBe background subtraction algorithm, which is implemented in this folder.
