MSc thesis subject: Object detection for automated airborne pest monitoring of Drosophila suzukii

Drosophila suzukii, also known as the spotted wing fruit fly, has become a serious pest in Europe attacking many soft-skinned crops such as several berry species and grapevines since its spread in 2008 to Spain and Italy. An efficient and accurate monitoring system to identify the presence of D. suzukii in crops and their surroundings is essential for the prevention of damage to economically valuable fruit crops.

Recent advancements in machine learning and deep learning in combination with object recognition allow for accurate detection of objects of interest. In this project, different algorithms for the detection of D. suzukii flies need to be tested. The challenging part is that the images in which the flies need to be detected will be taken from an unmanned aerial vehicle (UAV) in natural conditions, meaning different illumination conditions, un-sharp imagery due to movement of the UAV etc. Moreover, the suzukii flies are very small (±3 mm) and are very difficult to distinguish from other fruit fly species.

Existing methods for monitoring D. suzukii are costly, time and labour intensive, and typically conducted at a low spatial resolution. To overcome current monitoring limitations, we want to develop a novel system consisting of sticky traps which are monitored by means of UAVs and an image processing pipeline that automatically identifies and counts the number of D. suzukii per trap location. This thesis will be carry out the follow steps:

  • Collect images of  D. suzukii using UAVs and extend an existing database of training images;
  • Review of the state-of-the art of deep learning approaches for object/insect detection;
  • Test and implement different machine / deep learning methods for D. suzukii detection.


  • Cheng X, Zhang Y, Chen Y, Wu Y, Yue Y (2017). Pest identification via deep residual learning in complex background, Computers and Electronics in Agriculture, 141, 351-356.
  • Fuentes A, Yoon S, Kim S and Dong Park S (2017). A Robust Deep-Learning-Based Detector for Real-Time Tomato Plant Diseases and Pests Recognition, Sensors, 17, 2022.


  • UAV enthusiast
  • Willing to learn novel software and hardware tools
  • Excited to work in deep learning

Theme(s): Sensing & measuring