COS 86-1 - Deep artificial-intelligence learning for the Monarch butterfly instar identification and image classification

Abstract: With the rapid advancement of technology and participation of amateur naturalists, numerous images of insects are made available through web portals from around the globe. Scientific communities and conservationists are benefitting from these images for research and conservation purposes. Unfortunately, these images are often not easily interpretable, as they are not identified and classified by experts. To address this issue, we developed a computer-vision model utilizing the YOLOV5 algorithm to accurately identify and classify monarch butterfly instars, as an example, into the five developmental stages. The model was trained and tested using image data obtained from the iNaturalist citizen science program. It demonstrated a precision and accuracy of 93% in instar prediction and an overall accuracy of 85% for all five instar classes, with YOLOv5l version chosen for the best model performance and mAP scores ranging from 83% to 93%. The training data used in the study spanned from 2000 to March 2022, with the training photographs annotated and classified by university faculty experts. This model could be adopted in developing identification and classification models for various insects and pests in agricultural farms. As a follow-up, we are now working on developing a classification model to categorize various disease conditions of monarch and other butterfly instars. By employing this model, we can better equip scientific communities and conservationists for preparation of management plans for the present day and the future.