Development of machine learning and tailored image compression algorithms in bee hive monitoring systems

Abstract

Agriculture cannot function without beekeeping, which offers many advantages including honey production, pollination, and other bee-related products. However, it can be timeconsuming and difficult to manually check the effectiveness of bee colonies. We suggested a machine learning-based bee colony monitoring system that uses specialized image compression techniques to compress photos captured by cameras mounted in beehives. [1] The performance of the bee colony is then predicted using machine learning models that are applied to the compressed photos. To demonstrate the viability of the suggested strategy, we assessed the performance of the customized image compression algorithms and machine learning models and compared the outcomes with existing compression techniques and machine learning models. The methodology involved collecting datasets from open source DS-COMB-SEG-BEEHOPE and from Bee House Uganda Limited for validation, image segmentation using the UNet architecture, achieving an accuracy of 0.95, IoU of 0.89, and Dice score of 0.88, followed by cell detection using Circle Hough Transform with a detection rate of 0.987. Cell classification was performed using the MobileNet CNN model, attaining an accuracy of 0.92, precision of 0.90, recall of 0.88, and F1-Score of 0.89. An image compression algorithm that compresses JPEG image file formats was developed and a JPEG image of 12.1 MBs was compressed to 3.06 MBs with compression ration of 4: 1 , compression time is approximately 5 seconds and compression is 2.42 MBs per second