Skoči na glavni sadržaj

Stručni rad

https://doi.org/10.19279/TVZ.PD.2020-8-1-18

KLASIFIKACIJA AUTOMOBILA KORISTEĆI TRANSFERIRANO UČENJE NA RESNET ARHITEKTURI NEURONSKE MREŽE

Stjepan Ložnjak ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska, Student
Tin Kramberger ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska
Ivan Cesar ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska
Renata Kramberger ; Tehničko veleučilište u Zagrebu, Zagreb, Hrvatska


Puni tekst: engleski pdf 2.021 Kb

str. 59-64

preuzimanja: 344

citiraj


Sažetak

Classification is one of the most common problems that neural networks are used for. In the case of higher resolution image classification, convolutional neural networks are commonly used. Due to the reason that convolutional neural networks are so often used in classification, there are many pretrained models that can be adapted for new domains using a technique called transfer learning. This paper shows how excellent results in classification accuracy can be achieved by applying transfer learning to pretrained convolution neural network. This paper presents the results of the learning transfer of the ResNet-152 convolution neural network on the Stanford Cars dataset. The results show accuracy over 88% only by training the last fully connected layer.

Ključne riječi

Transfer learning; ResNet; Stanford Car dataset

Hrčak ID:

242768

URI

https://hrcak.srce.hr/242768

Datum izdavanja:

17.6.2020.

Podaci na drugim jezicima: hrvatski

Posjeta: 1.571 *