Optical insect identification using artificial intelligence: towards reliable in-field insect monitoring

Recent studies show that worldwide about 15% of crops are lost due to pest insects. Besides agriculture, the rapid identification of harmful insects is also of great importance for preventing the spread of human diseases such as malaria and zika. A sensor system that allows for the rapid identification of flying insects is currently lacking, and identification of insects is mainly based on counting insects on specialized traps. This procedure is subjective and time consuming. As a result, it is only performed on a weekly basis, and on a limited number of fields. Therefore, a rapid system for insect detection would be highly valuable. Recently, it was shown that flying insects can be identified based on their wingbeat characteristics. A first prototype sensor is available in the lab and should be further fine-tuned so that it can be used at farm level. The main goal of the PhD is to develop proper classification algorithms based on the optical wingbeat signatures. For this purpose, large datasets of millions of signatures are available through partners, and candidates should investigate the best data analysis techniques to process them. Concepts from time and frequency domain signal analysis will be used to pre-process the data. Classification algorithms borrowed from statistics, machine learning and deep learning disciplines are built and compared. The best model will then be implemented in a sensor and tested in the field. For those field trials, we will focus on Drosophila suzukii, a recent pest that hits the fruit production in Europe, and Napomyza cichorii, a pest insect causing important losses in the Belgian Endive sector. For these tests, we collaborate with research centers that are specialized in these insects.