Scale-Space Theory, F-transform Kernels and CNN Realization

被引：3

作者：

Molek, Vojtech ^{[1
]}

Perfilieva, Irina ^{[1
]}

机构：

[1] Univ Ostrava, Inst Res & Applicat Fuzzy Modeling, Ctr Excellence IT4Innovat, 30 Dubna 22, Ostrava, Czech Republic

来源：

ADVANCES IN COMPUTATIONAL INTELLIGENCE, IWANN 2019, PT II | 2019年 / 11507卷

关键词：

F-transform; Initialization; Scale-space; Convolutional neural network; Classification;

D O I：

10.1007/978-3-030-20518-8_4

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

We present scale-space and F-transform inspired modification to convolutional neural networks. The proposed modification improves network classification accuracy using multi-scale image representation and F-transform kernels pre-training. We evaluate our model on two databases and show better performance than networks without F-transform pre-training.

引用

页码：38 / 48

页数：11

共 28 条

[1]

[Anonymous], ARXIV151105641

[2]

[Anonymous], 2015, Nature, DOI [10.1038/nature14539, DOI 10.1038/NATURE14539]

[3]

[Anonymous], 2016, J FUZZY SET VALUED A, DOI DOI 10.5899/2016/JFSVA-00268

[4]

[Anonymous], 2014, P ADV NEUR INF PROC

[5]

[Anonymous], 2015, ARXIV151106856

[6]

Arora S, 2014, PR MACH LEARN RES, V32

[7]

Deng L., 2012, IEEE signal processing magazine, V29, P141, DOI [DOI 10.1109/MSP.2012.2211477, 10.1109/msp.2012.2211477]

[8]

Du Yunshu, 2017, ARXIV170904083

[9]

Erhan D, 2010, J MACH LEARN RES, V11, P625

[10]

Glorot X., 2010, P 13 INT C ART INT S, P249

← 1 2 3 →