Validation of convolutional layers in deep learning models to identify patterns in multispectral images
| Autor: | Isis Bonet, Fabio Caraffini, Alejandro Peña, Diego Manzur, Mario Gongora |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Artificial neural network
business.industry Computer science Deep learning Multispectral image Feature extraction 020207 software engineering Context (language use) Pattern recognition 02 engineering and technology Cartesian product Convolutional neural network symbols.namesake Computer Science::Computer Vision and Pattern Recognition 0202 electrical engineering electronic engineering information engineering symbols Feature (machine learning) 020201 artificial intelligence & image processing Artificial intelligence business |
| Popis: | The convolutional neural networks (CNN) are considered as a particular case of the Deep Learning neural networks, and have been widely used for the extraction of features in images, audio files or text recognition. For the automatic extraction of features from multispectral images, many researchers have appealed to the use of CNN models, which integrate layers with different structures in context with the solution of a problem, which suggests quite a challenge. That is why, in this article, we propose a method to evaluate the stability in the design of convolutional layers for labeling and identification of palm cultivation units from multispectral images. The structure of the proposed convolutional layer will be given in terms of a fuzzy feature map, obtained as a result of the Cartesian product of three vegetation indices commonly used to evaluate plant vigor in this type of crops (NDVI, GNDVI, RVI), represented as compact maps (radial basis functions). The stability in the design will be given in terms of the dominance of the main diagonal that defines the structure of a convolutional layer obtained as a result of the Cartesian product of two compact maps that represent the same multispectral image. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|