Inferring tephritid fly pupal development stage using non-invasive near infrared imaging and machine learning classification.

Abstract

Insect pupae change morphologically (e.g., pigmentation of eyes, wings, setae and legs) during the intrapuparial period. Knowledge on the physiological age of pupae and their emergence are important parameters for the control of agriculturally important Tephritid flies. Traditional methods for determining age require dissecting the puparium, thus killing the specimen. Therefore, non-invasive and more ethical methods to determine physiological age are needed, especially if individual pupae are followed throughout their development. Furthermore, machine learning methods can be employed to detect pupal age, thereby reducing human-bias. Here, we studied the intrapuparial development of the Mexican fruit fly, Anastrepha ludens (Diptera: Tephritidae), using non-invasive near-infrared (NIR, 850-1100 nm) images. We photographed pupae and subsequently analysed the images with machine learning algorithms. The intrapuparial period lasted between 17 and 19 days at a constant temperature of 26°C, and 75-80% relative humidity. No visible structures were observed between days 1 and 3. The phanerocephalic pupa was observed on day 4. The darkening of the eyes began on day 12. Wing pigmentation occurred on days 13 and 14, and the legs and setae on the thorax became melanized on day 15. A convolutional neural network correctly identified the physiological age range of intrapuparial development stages with an average accuracy of 71.77%. This model using NIR imaging allows the determination of a physiological age range without arresting the development of the pupae, and an estimation of the viability of pupae without waiting for the emergence of the adult.