Number of insect classes in the dataset used for training, validation and test.

Insects were annotated from the images with Sedum plants. The background class contains images without any insects. A train and validation dataset with six species (6cls) was used to train the first YOLO model. A benchmark training and validation dataset with an approximate split percentages of 85/15 with nine species (9cls) was used to train the final evaluated YOLO models. The test component of the benchmark dataset contains additional unfamiliar classes 10–19 that contain other species of bees, flies and beetles, unclear insects and other animals. Classes 1–9 are listed in brackets next to their closest relative of the unfamiliar classes 10–19.