An annotated image dataset of medically and forensically important flies for deep learning model training

Abstract

Conventional methods to study insect taxonomy especially forensic and medical dipterous flies are often tedious, time-consuming, labor-intensive, and expensive. An automated recognition system with image processing and computer vision provides an excellent solution to assist the process of insect identification. However, to the best of our knowledge, an image dataset that describes these dipterous flies is not available. Therefore, this paper introduces a new image dataset that is suitable for training and evaluation of a recognition system involved in identifying the forensic and medical importance of dipterous flies. The dataset consists of a total of 2876 images, in the input dimension (224 × 224 pixels) or as an embedded image model (96 × 96 pixels) for microcontrollers. There are three families (Calliphoridae, Sarcophagidae, Rhiniidae) and five genera (Chrysomya, Lucilia, Sarcophaga, Rhiniinae, Stomorhina), and each class of genus contained five different variants (same species) of fly to cover the variation of a species.