Segmentação de áreas de interesse em quadros de cria de abelhas por meio de visão computacional

Abstract

Machine learning is an area that has been growing and taking up more and more space in both mundane and complex day-to-day processes, in the beekeeping area this is no different, with recent promising research being done using Semantic Segmentation of images for analysis of bees honeycomb, this study brings an analysis and evaluation using the Intersect over Union method in different types of neural network architectures and encoders for segmentation of Apis Mellifera honeycombs. This work carried out a comparative study of semantic segmentation techniques and from this it was possible to verify an increase in the state-of-the-art, with the use of the SegNet architecture associated with the MobileNet encoder for feature extraction. Based on the results obtained, it is suggested to replace the originally proposed model, aiming to improve the results of the combs classification pipeline.