Design and Evaluation Support System for Convolutional Neural Network, Support Vector Machine and Convolutional Autoencoder

Fifth Author's Department

Robotics, Control & Smart Systems Program

Find in your Library

https://doi.org/10.1201/9781003343783-3

All Authors

Fusaomi Nagata, Kento Nakashima, Kohei Miki, Koki Arima, Tatsuki Shimizu, Keigo Watanabe, Maki K. Habib

Document Type

Research Article

Publication Title

Measurements and Instrumentation for Machine Vision

Publication Date

1-1-2024

doi

10.1201/9781003343783-3

Abstract

Recently, deep learning approaches such as Convolutional Neural Network (CNN), Support Vector Machine (SVM), convolutional autoencoders (CAE) are widely applied to solve various kinds of industrial defect detection problems. Since 2017, the authors have developed a design, training, and evaluation support software available on MATLAB® for CNN, SVM, CAE, etc. Figure 3.1 shows the main dialogue of the developed application. Original models designed through the application can be applied to various classification and anomaly detection problems. As for CNN, not only initial learning-based CNN (ILCNN) with an original series-type structure but also the transfer learning-based CNN (TLCNN) using the convolutional blocks of already trained large-scaled CNNs such as AlexNet, GoogLeNet, VGG16, VGG19 and so on can be designed. As for SVM, two types of models can be constructed.

First Page

66

Last Page

82

This document is currently not available here.

Find this item online

Share

COinS