Publicação
Towards precise recognition of pollen bearing bees by convolutional neural networks
| Resumo: | Automatic recognition of pollen bearing bees can provide important information both for pollination monitoring and for assessing the health and strength of bee colonies, with the consequent impact on people's lives, due to the role of bees in the pollination of many plant species. In this paper, we analyse some of the Convolutional Neural Networks (CNN) methods for detection of pollen bearing bees in images obtained at hive entrance. In order to show the in uence of colour we preprocessed the dataset images. Studying the results of nine state-of-the-art CNNs, we provide a baseline for pollen bearing bees recognition based in deep learning. For some CNNs the best results were achieved with the original images. However, our experiments showed evidence that DarkNet53 and VGG16 have superior performance against the other CNNs tested, with unsharp masking preprocessed images, achieving accuracy results of 99:1% and 98:6%, respectively. |
|---|---|
| Autores principais: | Monteiro, Fernando C. |
| Outros Autores: | Pinto, Cristina M.; Rufino, José |
| Assunto: | Pollen bearing bees Convolutional neural network Deep learning |
| Ano: | 2021 |
| País: | Portugal |
| Tipo de documento: | comunicação em conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Politécnica de Bragança |
| Idioma: | inglês |
| Origem: | Biblioteca Digital da UPB |
| Resumo: | Automatic recognition of pollen bearing bees can provide important information both for pollination monitoring and for assessing the health and strength of bee colonies, with the consequent impact on people's lives, due to the role of bees in the pollination of many plant species. In this paper, we analyse some of the Convolutional Neural Networks (CNN) methods for detection of pollen bearing bees in images obtained at hive entrance. In order to show the in uence of colour we preprocessed the dataset images. Studying the results of nine state-of-the-art CNNs, we provide a baseline for pollen bearing bees recognition based in deep learning. For some CNNs the best results were achieved with the original images. However, our experiments showed evidence that DarkNet53 and VGG16 have superior performance against the other CNNs tested, with unsharp masking preprocessed images, achieving accuracy results of 99:1% and 98:6%, respectively. |
|---|