Deep Semantic Segmentation Applied to Honey Bee Comb Cells

• Gustavo Vasconcelos UTFPR
• André Roberto Ortoncelli UTFPR https://orcid.org/0000-0001-9622-8525
• Fabiana Martins Costa UTFPR
• Marlon Marcon UTFPR https://orcid.org/0000-0002-3698-8570

Resumo

Machine learning is increasingly prevalent in various fields, including beekeeping, where it has been applied to honeycomb analysis through Semantic Segmentation of images. This study evaluates different neural network architectures and encoders using the Intersection over Union (IoU) method to segment Apis mellifera honeycombs. A comparative analysis of semantic segmentation techniques was conducted, revealing advancements in the field with the SegNet architecture combined with the MobileNet encoder for feature extraction. We present a new dataset of 10 comb cell images, which we combine with the images from the DeepBee dataset in our experiments. Our results surpass the state-of-the-art results for comb cell segmentation in the DeepBee dataset.

Palavras-chave: Semantic Segmentation, Honeycombs, Convolutional Neural Network, Encoders