USING AI TO FIGHT INVASIVE SPECIES

The Yellow Legged Asian hornet (*Vespa velutina*) is a highly predative non-native species of hornet that hails from East Asia and has spread rapidly throughout Europe after being inadvertently introduced in 2004.

The presence of Asian Hornets in the UK needs to be taken seriously due to their severe threat to our ecological security as well as their impact on food sources . Experience from Europe show us that the management costs to this country could exceed £7million a year. Since 2016, a surge in reported sightings has overwhelmed expert teams reviewing them, with 99% turning out to be misidentifications.

We've partnered with Capgemini's Applied Innovation Exchange (AIE) in London to develop a companion smartphone app featuring AI-powered AH identification. This user-friendly app, working seamlessly with the our Asian Hornet Bait Station, Nest Sweeper , allowing for swift and accurate identification of these invasive hornets. With object detection computer vision models the technology can detect and monitor the presence of Asian hornets. It provides us with real-time information, enabling the team, and the relevant authorities to take immediate defensive action.

DOWNLOAD THE APP

JOURNEY OF HORNET AI

We've spent the last few years studying AH behaviour, including multiple research trips to Jersey, a hot-spot for Asian Hornet activity, with the invaluable support of Alastair Christie BSc (Senior Scientific Officer for invasive species for the Government of Jersey). Having recognised the severity of the threat, our Co-Founder Matthew has been tirelessly pioneering innovative solutions aimed at addressing the imminent threat of this insect spreading throughout the UK mainland.

Before Hornet AI technology became ‘app ready’, Matthew worked on prototypes and hired Ver Facil Limited in Bodmin, a specialist in remote monitoring solutions, to connect the low-power, long-range wireless LoRaWAN network to the alert system. This system allowed the devices to work without internet connection. To train the model, specialist annotation software from Encord was used. It trained a robust model with rectangular boxes on 1,500 images for each of the following types of insects: Vespa velutina, Vespa crabro, Vespula vulgaris, and Apis mellifera. These images were obtained from iNaturalist, and a faster R-CNN model was trained. Faster R-CNN is an improved version of RCNN (Region-based Convolutional Neural Network).