Researchers are able to detect fruit flies in images of sticky traps using object detection and deep learning.
The fruit fly, spotted wing drosophila, is a pest found worldwide that attacks many soft-skinned fruit, and has led to dramatic loses for farmers. Its monitoring and management currently relies on the use of liquid bait traps and manual counting, a costly and time consuming process.
A team of researchers from across Europe have collaborated to investigate the use of image-based object detection with deep learning to monitor insect traps. Computer vision-based insect detection has received more attention due to advances in the field of object detection and deep learning, especially with the use of convolutional neural networks, a class of deep neural network found to be effective in image recognition and classification.
The researchers, who reported their findings in SCI journal Pest Management Science, investigated the possibility of using deep learning for monitoring of insect traps under field conditions and the potential of insect detection in images collected from unmanned aerial vehicle (UAV). Both data sets are required outdoors under varying illumination conditions.
Researchers found that that it was possible to detect fruit flies in images of sticky traps using object detection and deep learning. They were also able to differentiate between male and female flies as well as specific fruit fly species. The researchers added that monitoring pest insects using deep learning object detection methods is a niche area of computer vision and therefore assessment of model performance is not standardised. It was also noted that a custom UAV system with a high grade camera was needed as off-the-shelf UAV systems are not able to collect images of high enough quality for detecting objects as small as the insects the researchers were interested in. The researchers anticipate that having developed a proof of concept this computer vision-based strategy for managing fruit flies could be applied more widely.