| |
|
|
| Registros recuperados : 96 | |
| Registros recuperados : 96 | |
|
|
|
 | Acesso ao texto completo restrito à biblioteca da Embrapa Agricultura Digital. Para informações adicionais entre em contato com cnptia.biblioteca@embrapa.br. |
Registro Completo
|
Biblioteca(s): |
Embrapa Agricultura Digital. |
|
Data corrente: |
03/05/2019 |
|
Data da última atualização: |
07/01/2020 |
|
Tipo da produção científica: |
Artigo em Periódico Indexado |
|
Circulação/Nível: |
A - 1 |
|
Autoria: |
BARBEDO, J. G. A.; CASTRO, G. B. |
|
Afiliação: |
JAYME GARCIA ARNAL BARBEDO, CNPTIA; GUILHERME B. CASTRO, CromAI, São Paulo. |
|
Título: |
Influence of image quality on the identification of psyllids using convolutional neural networks. |
|
Ano de publicação: |
2019 |
|
Fonte/Imprenta: |
Biosystems Engineering, v. 182, p. 151-158, 2019. |
|
DOI: |
https://doi.org/10.1016/j.biosystemseng.2019.04.007 |
|
Idioma: |
Inglês |
|
Conteúdo: |
Convolutional Neural Networks (CNNs) usually require large datasets to be properly trained. Although techniques such as transfer learning can relax those requirements, gathering sufficient labelled data to cover all the variability associated to the problem at hand is often costly and time consuming. A way to minimise this challenge would be gathering the training data under laboratory conditions, using high quality sensors capable of generating images with superior resolution, sharpness and contrast. The downside of this approach is that the resulting dataset will most likely lack the variety that can be found under more realistic conditions. This work investigates this trade-off between image quality and dataset representativeness, that is, if a CNN trained with images captured by a scanner in laboratory would be able to reliably recognise psyllids in smartphone images captured under more realistic conditions. A total of 1276 images were used in the experiments, half acquired using a flatbed scanner and half acquired using two different brands of smartphones. Experiments were carried out using Squeezenet CNNs and a 10-fold crossvalidation strategy. Accuracies ranged from less than 70% using only scanned images, to around 90% when only smartphone images were employed, indicating that more realistic conditions are essential to guarantee the robustness of the trained network. Scanned images were useful when the training set containing realistic images was not enough to cover all the variability found in the experiments, but were otherwise innocuous. MenosConvolutional Neural Networks (CNNs) usually require large datasets to be properly trained. Although techniques such as transfer learning can relax those requirements, gathering sufficient labelled data to cover all the variability associated to the problem at hand is often costly and time consuming. A way to minimise this challenge would be gathering the training data under laboratory conditions, using high quality sensors capable of generating images with superior resolution, sharpness and contrast. The downside of this approach is that the resulting dataset will most likely lack the variety that can be found under more realistic conditions. This work investigates this trade-off between image quality and dataset representativeness, that is, if a CNN trained with images captured by a scanner in laboratory would be able to reliably recognise psyllids in smartphone images captured under more realistic conditions. A total of 1276 images were used in the experiments, half acquired using a flatbed scanner and half acquired using two different brands of smartphones. Experiments were carried out using Squeezenet CNNs and a 10-fold crossvalidation strategy. Accuracies ranged from less than 70% using only scanned images, to around 90% when only smartphone images were employed, indicating that more realistic conditions are essential to guarantee the robustness of the trained network. Scanned images were useful when the training set containing realistic images was not enough to cover ... Mostrar Tudo |
|
Palavras-Chave: |
Convolutional neural network; Deep learning; Image processing; Processamento de imagem; Psyllids; Redes neurais. |
|
Thesaurus NAL: |
Image analysis. |
|
Categoria do assunto: |
X Pesquisa, Tecnologia e Engenharia |
|
Marc: |
LEADER 02308naa a2200229 a 4500
001 2108668
005 2020-01-07
008 2019 bl uuuu u00u1 u #d
024 7 $ahttps://doi.org/10.1016/j.biosystemseng.2019.04.007$2DOI
100 1 $aBARBEDO, J. G. A.
245 $aInfluence of image quality on the identification of psyllids using convolutional neural networks.$h[electronic resource]
260 $c2019
520 $aConvolutional Neural Networks (CNNs) usually require large datasets to be properly trained. Although techniques such as transfer learning can relax those requirements, gathering sufficient labelled data to cover all the variability associated to the problem at hand is often costly and time consuming. A way to minimise this challenge would be gathering the training data under laboratory conditions, using high quality sensors capable of generating images with superior resolution, sharpness and contrast. The downside of this approach is that the resulting dataset will most likely lack the variety that can be found under more realistic conditions. This work investigates this trade-off between image quality and dataset representativeness, that is, if a CNN trained with images captured by a scanner in laboratory would be able to reliably recognise psyllids in smartphone images captured under more realistic conditions. A total of 1276 images were used in the experiments, half acquired using a flatbed scanner and half acquired using two different brands of smartphones. Experiments were carried out using Squeezenet CNNs and a 10-fold crossvalidation strategy. Accuracies ranged from less than 70% using only scanned images, to around 90% when only smartphone images were employed, indicating that more realistic conditions are essential to guarantee the robustness of the trained network. Scanned images were useful when the training set containing realistic images was not enough to cover all the variability found in the experiments, but were otherwise innocuous.
650 $aImage analysis
653 $aConvolutional neural network
653 $aDeep learning
653 $aImage processing
653 $aProcessamento de imagem
653 $aPsyllids
653 $aRedes neurais
700 1 $aCASTRO, G. B.
773 $tBiosystems Engineering$gv. 182, p. 151-158, 2019.
Download
Esconder MarcMostrar Marc Completo |
|
Registro original: |
Embrapa Agricultura Digital (CNPTIA) |
|
|
Biblioteca |
ID |
Origem |
Tipo/Formato |
Classificação |
Cutter |
Registro |
Volume |
Status |
Fechar
|
| Nenhum registro encontrado para a expressão de busca informada. |
|
|
