A REAL-TIME FRAMEWORK FOR MONITORING AND ANALYSING BEE BEHAVIOUR DURING POLLEN FORAGING

Abstract

Increases in temperature and light intensity led to a rise in the quantity of forag- ing excursions for pollen, while increases in relative humidity, precipitation, and wind speed resulted in a decrease. When there was a lot of rain or a light breeze, there was a noticeable decrease in pollen foraging activity. This study shows the effectiveness of the suggested imaging technology in addition to quantitatively presenting the impact of environmental conditions on pollen foraging behavior. The imaging system can be mechanizedused as a dependable and effective tool to help beekeepers manage their hives and to help researchers better understand the foraging behavior of honey bees. The maintenance of a healthy ecosystem depends heavily on honeybees. In addition to producing honey, beeswax, and royal jelly, bees also aid in pollination. Although beekeeping has a long history and signifi- cant global economic potential, swarming is still a major obstacle to sustaining profitability. Since swarming reduces the number of bees in hives and affects honey output, it has a major effect on beekeepers’ profitability. Swarming is a common colony reproductive process in honeybees. Therefore, it is crucial to monitor these beehives in order to keep an eye on their erratic behavior. Visual inspection of hives, temperature monitoring, or machine learning analysis of auditory data can all be used to anticipate swarms. Since acoustic monitoring circumvents the limitations of visual examinations and is unaffected by environmental vari- ables like temperature, it is essential for identifying changes in colony behavior. A mixed machine learning approach to bee behavior analysis that prioritizes spatial detection and a temporal activity classification. Our method enables efficient real-time monitoring and clas- sification of bee activity, yielding valuable data for ecological research and precision farming. The system exhibits significant improvements in behavioral classification and precision in detection when evaluated on a custom-labeled video dataset. We study techniques such as Multi-Layer Perceptron (MLP) and Logistic Regression to address the issue of identifying a queen bee in a hive. MFCC are utilized for feature extraction and as input for models of classification in order to distinguish between hives with and without a queen bee.