Short Hive Audio Recording Predictor

Over the past decade, researchers and practitioners have explored the potential of audio recording for precision beekeeping. Most of the research focused on in – hive sensors and real – time detection of critical events, such as swarming, queen state or hive strength. However, adoption of such sensors is difficult in a low margin industry such as beekeeping. On the other hand, most, if not all, beekeepers have a smartphone. As such, in this work we explore the potential of using short audio segments recorded from beehives. Such non – invasive assessment has been gaining interest mainly to allow beekeepers to lower the number of interventions they make on a hive.

The project aims to extend the model output and optimize its performance. It will developing a Federated learning logic. Many AI systems leverage scale to reach performance, leading to huge energy consumption to accomplish their jobs. At Nectar, we promote a sustainable farming system and building such a system would be counterproductive. Additionally, there is a variation in how beekeepers report strength, mite level and queen status. By personalizing the model on device, we can better personalize the predictions to match the user behavior. Finally, our users often work remotely with low internet connections and our system needs to work offline, therefore making the inference on the device is core to our system.

Developing and deploying such a tool could truly change the paradigm of beekeeping. Indeed, the main driver of mortality is the lack of proper management of varroa destructors, the main pest in beekeeping. By allowing users to not only disrupt the bees in general but also sample quickly and often the mite level, we can promote a true IPM program and solve the biggest challenge of beekeeping, managing mites ! Honeybee pollinate 33% of the food we eat, and the average mortality rate of the last 3 years in the US is over 50% putting our food system at risk and by nature putting our society at risk.

Faculty Supervisor:

Tiago Henrique Falk

Student:

Partner:

Nectar Technologies

Discipline:

Computer science

Sector:

Agriculture

University:

Université du Québec : Institut national de la recherche scientifique

Program: