From page 37 significant factor is the cost of these scales , with some commercial scales exceeding $ 1,000 per unit . The majority of Australian beekeepers cannot afford a scale for each hive and this calls for an unconventional solution for this important problem . The experiments carried out during this PhD project , aided by the previous research on honeybees , identified several parameters that either dictate or indicate honey bee foraging activity . For example , the weather at the hive site dictates the time honeybees can spend foraging ; colony size determines the number of foragers which leave the hive to find pollen and nectar . The thermoregulation inside a beehive is dependent upon the size of the bee colony so together with humidity and CO2 levels inside the hive , can be used to differentiate between colonies of different sizes . It has been observed that , despite similar weather conditions and colony size , some colonies gain more weight than the others . Presence of a healthy queen , bee genetics , past nutritional status , and the ability of foragers to find good resources , make a significant difference . This is where the bee acoustics inside the hive can be used to determine how active the colony is , with the accelerometer picking up vibrations caused by the waggle dance that leads the colony to access foraging resources .
Using artificial intelligence
Instead of using the expensive and bulky weighing scales to directly measure the weight of the hive , this project used cutting-edge Artificial Intelligence ( AI ), data from cost effective sensors inside the hive , weather data , and the seasonal information to estimate the daily change in hive weight . Figure 3 shows the weight estimation process , where important features are extracted from the collected data , and a deep network trained on these features generates the daily weight variation estimates . The actual weight of hive recorded using the weighing scale is used to determine the accuracy of the estimates , and improve the AI . The
Figure 3 : The design of beehive weight estimation system .
Figure 4 : Long term weight tracking of a hive .
weight estimation results for hive data collected over a year , using eight systems and various sites across south-west WA show promising accuracies . These daily weight estimates can be stitched together to estimate the weight of a hive over long period of time . Figure 4 shows the actual weight of a hive compared to the estimated weight for over 300 days . The sharp changes in the hive weight are a result of hive manipulations carried out by the beekeeper . It is a routine practice to add or remove supers from the hive based on season and colony strength . The AI needs an updated weight after such manipulations to continue tracking the hive weight .
However , between such manipulations , the AI is able to track the hive weight on its own with good accuracy . This allows the beekeepers to assess the honey bee hive health and foraging activity without the expensive weighing scales . To translate this work into a commercially available beehive monitoring system , there is still work to be done . A lot more data acquisition systems need to be developed and deployed across Australia , especially in hives that are used for the pollination of almonds . The data collected from these systems will be used to train the AI models for weight estimation . Modifications will be required to accommodate different regions and weather patterns . With adequate data to train the AI , this system has the potential to estimate the beehive weight in all conditions , thus providing a measure of pollination carried out by each hive in a very cost-effective manner .
The technical details of this work have been published by authors in AAAI Conference 2022 . If you would like to assist in the continued road to commercialisation , please contact Omar Anwar ( email :
o _ anwar @ ymail . com ) or Y-Trace ( email :
ytrace2022 @ gmail . com )
38 In A Nutshell - Spring 2022 Vol 23 Issue 3