Effect of the SVM algorithm on deep classification networks

Abstract

The study aimed at applying several convolutional neural networks on two data samples containing a large group of human images in order to identify them through transfer learning technology, and study the effect of applying support vector machine (SVM) on these networks.

Human facial recognition technology is an important problem; this technology is applied broadly in security (such as criminal identification), education (such as attendance systems), protection systems (such as secure electronic banking), etc.

Traditional algorithms didn't give optimal results in the classification field, so we used the newest (CNNs) in its.

Additionally, we replaced classification layer in each of the networks studied with (svm), to study its effect on the performance of these networks in accuracy and time.

At the end, we got good results that achieved accuracy about 99% and reducing training time and classification error rate in some cases.