Artificial intelligence achieves a key step in the management of fruit flies


The Mexican fruit fly (Anastrepha ludens) is one of the most harmful insect pests in the world. A key method of managing them is the sterile insect technique, in which sterile male flies are mass-reared and released into the wild, after which they mate with wild females, which then produce no offspring. Determining the precise age of mass-reared fruit flies is a critical step in the sterile insect technique, and Mexican researchers have applied machine learning algorithms that can accurately measure the age of fruit fly pupae in fruit to properly time the irradiation. (Photo by Andrés Diaz Cervantes)

By Diana Pérez-Staples, Ph.D., and Horacio Tapia-McClung, Ph.D.

Horacio Tapia-McClung, Ph.D.

Horacio Tapia-McClung, Ph.D.

Diana Perez-Staples, Ph.D.

Diana Perez-Staples, Ph.D.

Two of the most damaging pests in the world are the Mediterranean fruit fly (Ceratitis capitata) and the Mexican fruit fly (Anastrepha ludens), causing billions of dollars in damage to agriculture. Fortunately, the sterile insect technique is currently being used as part of integrated management programs on a regional scale to control these flies in certain regions of the world.

The Sterile Insect Technique (ITS) is a type of birth control, involving raising millions of these flies in factories, irradiating them with X-rays or gamma rays to render them sterile, and then releasing them into areas where pests are present. When sterile males mate with wild females, the females will not have fertile eggs to lay in fruit. Thus, the population levels are decreased. SIT has good ecological credentials because it targets only pest species, does not introduce foreign genetic material into the population, and reduces the use of insecticides.

The irradiation process in SIT is the key to its success. For tephritid flies, irradiation is usually carried out a few days before the pupae emerge as adults. If pupae are irradiated too early or too late in their development process, it can lead to mobility and behavior problems in adulthood. However, even under controlled conditions, pupae can vary in their development time. Thus, one of the tests performed before irradiation is to determine the physiological age of the pupae.

Currently, in these fruit fly factories around the world, technicians must determine the correct time to irradiate by taking a sample of the pupae, removing the pupal case to expose the eyes, and then checking the color of the eyes against to a color chart. This can be laborious and prone to human error, as it depends on the skill, experience and expertise of the technician, as well as natural biases in color interpretation. Technicians can be fatigued by this repetitive work, while sick leave and vision problems can also cause variations in the correct determination.

Artificial intelligence to the rescue

At Universidad Veracruzana, in collaboration with the Secretary of Agriculture of Mexico (Programa Operativo de Moscas, DGSV-SENASICA), we have partnered with experts in artificial intelligence to develop methods based on algorithms capable of determining with accurately age a pupa from a digital image captured with a common mobile device. We share our results in a new article published this month in the Journal of Economic Entomology.

Ivan Gonzalez-Lopez

Ivan Gonzalez-Lopez

For this, and as part of his doctorate. at the Facultad de Ciencias Agrícolas of the Universidad Veracruzana, Iván González-López, currently based at the IAEA-FAO Entomology Laboratory in Austria, took photographs of the exposed eyes of Mediterranean fruit fly nymphs and Mexican fruit flies . We chose pupae that still had a few days to emerge and deliberately took rough pictures that didn’t have perfect lighting or focus conditions. In fact, they were caught quickly and with a cell phone.

Then, as part of her master’s research at the Laboratorio Nacional de Informática Avanzada in Xalapa Veracruz, Georgina Carrasco processed the images with a program trained to detect the eye area in the photograph and crop it. Then, using the correct answers from a plant technician, another algorithm was trained through a supervised machine learning method known as transfer learning, to accurately determine the age of the pupae.

We found that algorithms based on a neural network architecture known as Inception v1 correctly identified the physiological age of maturity two days before emergence with 75% accuracy for the Mexican fruit fly and 83.16% for the Mediterranean fruit fly, respectively. This method isn’t perfect for sure, and it still requires a technician to dissect the pupae and take pictures, but it’s a promising approximation of how supervised machine learning and artificial intelligence can be used. to help with uncertainty in decisions about when to irradiate. The level of accuracy can also be improved as more photos are taken and provided to the algorithm to learn from.

The next steps will be to develop software that could easily be used by technicians as well as to train these algorithms with other tephritidae pest species currently controlled by SIT. Certainly, this highlights that there can be exciting collaborations between entomologists and artificial intelligence researchers.

Diana Perez-Staples, Ph.D., is a professor-researcher at the Institute of Biotechnology and Applied Ecology of the Universidad Veracruzana, in Xalapa, Veracruz, Mexico. E-mail: [email protected]. Horacio Tapia-McClung, Ph.D., is a professor-researcher at the Artificial Intelligence Research Institute of the Veracruzana University also in Xalapa. E-mail: [email protected].


About Author

Comments are closed.