A Comparison of the Accuracies of a Convolution Neural Network Built on Different Types of Convolution Layers

Abstract

The development of artificial intelligence aims to increase its reliability, which is determined ambiguously by accuracy. The accuracy increase is key conception in DNN and AI at all. There is a strong relation between training the DNN and accuracy so the training is very important and very complicated. For that reason, looking for high accuracy in fire recognition it is made research a part of which will present in this article. We presented Convolution Neural Network (CNN) created by Python with Keras library can be very useful in early forest fire detection. To improve the accuracy is made private database with fire and smoke pictures exclusively. We tested two CNN architectures and Batch normalization in these architectures. The distribution of each layer's inputs change during training as the parameters of previous layer changed make the training mode complicated. Without claiming a single indicator, the training mode is expressed by trainable and non-trainable parameters related with Batch normalization layer in CNN. The influence of Batch normalization on accuracy are presented in tabular form. The accuracy comparison of two architectures with different layers in CNN is presented in tabular form.