A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures
Some recent studies show that filters in convolutional neural networks (CNNs) have low color selectivity in datasets of natural scenes such as Imagenet. CNNs, bio-inspired by the visual cortex, are characterized by their hierarchical learning structure which appears to gradually transform the repres...
| Главные авторы: | , , , |
|---|---|
| Формат: | Статья |
| Язык: | English |
| Опубликовано: |
MDPI AG
2023-08-01
|
| Серии: | Sensors |
| Предметы: | |
| Online-ссылка: | https://www.mdpi.com/1424-8220/23/17/7582 |
| _version_ | 1827727761032085504 |
|---|---|
| author | Oscar Sanchez-Cesteros Mariano Rincon Margarita Bachiller Sonia Valladares-Rodriguez |
| author_facet | Oscar Sanchez-Cesteros Mariano Rincon Margarita Bachiller Sonia Valladares-Rodriguez |
| author_sort | Oscar Sanchez-Cesteros |
| collection | DOAJ |
| description | Some recent studies show that filters in convolutional neural networks (CNNs) have low color selectivity in datasets of natural scenes such as Imagenet. CNNs, bio-inspired by the visual cortex, are characterized by their hierarchical learning structure which appears to gradually transform the representation space. Inspired by the direct connection between the LGN and V4, which allows V4 to handle low-level information closer to the trichromatic input in addition to processed information that comes from V2/V3, we propose the addition of a long skip connection (LSC) between the first and last blocks of the feature extraction stage to allow deeper parts of the network to receive information from shallower layers. This type of connection improves classification accuracy by combining simple-visual and complex-abstract features to create more color-selective ones. We have applied this strategy to classic CNN architectures and quantitatively and qualitatively analyzed the improvement in accuracy while focusing on color selectivity. The results show that, in general, skip connections improve accuracy, but LSC improves it even more and enhances the color selectivity of the original CNN architectures. As a side result, we propose a new color representation procedure for organizing and filtering feature maps, making their visualization more manageable for qualitative color selectivity analysis. |
| first_indexed | 2024-03-10T23:12:32Z |
| format | Article |
| id | doaj.art-a4bde274b51e4d4fb7ed05a15bb3c31b |
| institution | Directory Open Access Journal |
| issn | 1424-8220 |
| language | English |
| last_indexed | 2024-03-10T23:12:32Z |
| publishDate | 2023-08-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Sensors |
| spelling | doaj.art-a4bde274b51e4d4fb7ed05a15bb3c31b2023-11-19T08:51:47ZengMDPI AGSensors1424-82202023-08-012317758210.3390/s23177582A Long Skip Connection for Enhanced Color Selectivity in CNN ArchitecturesOscar Sanchez-Cesteros0Mariano Rincon1Margarita Bachiller2Sonia Valladares-Rodriguez3Department of Artificial Intelligence, National University of Distance Education (UNED), 28040 Madrid, SpainDepartment of Artificial Intelligence, National University of Distance Education (UNED), 28040 Madrid, SpainDepartment of Artificial Intelligence, National University of Distance Education (UNED), 28040 Madrid, SpainDepartment of Artificial Intelligence, National University of Distance Education (UNED), 28040 Madrid, SpainSome recent studies show that filters in convolutional neural networks (CNNs) have low color selectivity in datasets of natural scenes such as Imagenet. CNNs, bio-inspired by the visual cortex, are characterized by their hierarchical learning structure which appears to gradually transform the representation space. Inspired by the direct connection between the LGN and V4, which allows V4 to handle low-level information closer to the trichromatic input in addition to processed information that comes from V2/V3, we propose the addition of a long skip connection (LSC) between the first and last blocks of the feature extraction stage to allow deeper parts of the network to receive information from shallower layers. This type of connection improves classification accuracy by combining simple-visual and complex-abstract features to create more color-selective ones. We have applied this strategy to classic CNN architectures and quantitatively and qualitatively analyzed the improvement in accuracy while focusing on color selectivity. The results show that, in general, skip connections improve accuracy, but LSC improves it even more and enhances the color selectivity of the original CNN architectures. As a side result, we propose a new color representation procedure for organizing and filtering feature maps, making their visualization more manageable for qualitative color selectivity analysis.https://www.mdpi.com/1424-8220/23/17/7582color selectivityskip connectionslong skip connectionCNNVGG16Densenet121 |
| spellingShingle | Oscar Sanchez-Cesteros Mariano Rincon Margarita Bachiller Sonia Valladares-Rodriguez A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures Sensors color selectivity skip connections long skip connection CNN VGG16 Densenet121 |
| title | A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures |
| title_full | A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures |
| title_fullStr | A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures |
| title_full_unstemmed | A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures |
| title_short | A Long Skip Connection for Enhanced Color Selectivity in CNN Architectures |
| title_sort | long skip connection for enhanced color selectivity in cnn architectures |
| topic | color selectivity skip connections long skip connection CNN VGG16 Densenet121 |
| url | https://www.mdpi.com/1424-8220/23/17/7582 |
| work_keys_str_mv | AT oscarsanchezcesteros alongskipconnectionforenhancedcolorselectivityincnnarchitectures AT marianorincon alongskipconnectionforenhancedcolorselectivityincnnarchitectures AT margaritabachiller alongskipconnectionforenhancedcolorselectivityincnnarchitectures AT soniavalladaresrodriguez alongskipconnectionforenhancedcolorselectivityincnnarchitectures AT oscarsanchezcesteros longskipconnectionforenhancedcolorselectivityincnnarchitectures AT marianorincon longskipconnectionforenhancedcolorselectivityincnnarchitectures AT margaritabachiller longskipconnectionforenhancedcolorselectivityincnnarchitectures AT soniavalladaresrodriguez longskipconnectionforenhancedcolorselectivityincnnarchitectures |