Development of machine learning and tailored image compression algorithms in beehive monitoring system

Abstract

Beekeeping is a vital component of agriculture, providing numerous benefits such as pollination, honey production, and other bee-related products. However, monitoring the health and performance of bee colonies can be time-consuming and laborious. We proposed a machine learning-based monitoring system for bee colonies that uses tailored image compression algorithms to compress images taken using cameras installed in beehives. The compressed images are then analyzed using machine learning models to predict bee colony performance. [1] We evaluated the performance of the tailored image compression algorithms and machine learning models and compare the results with other compression algorithms and machine learning models to demonstrate the effectiveness of the proposed approach. We developed a tailored image compression algorithm that could compress JPG image formats from with compression time of 5 seconds, compression ratio of 4.1 and compression speed of 2.42 MBs per second We developed an image segmentation model based on the UNet architecture, trained on the DS-COMB-SEG-BEEHOPE dataset, achieving remarkable accuracy, IoU, and Dice scores of 0.95, 0.89, and 0.88 respectively. The model accurately delineated the boundaries of comb cells, providing precise masks for further analysis. By combining the power of UNet-based segmentation and Circle Hough Transform, we achieved an outstanding cell detection rate of 0.987. This approach significantly improved the accuracy of identifying comb cells, allowing for reliable analysis and subsequent classification using the MobileNet CNN model. The cell classification model demonstrated an impressive accuracy of 0.92, along with precision, recall, and F1-Scores of 0.90, 0.88, and 0.89 respectively, on predicting the compressed images we obtained an error of 0.13 percent.