February 29, 2024
Amin Torabi Jahromi

Amin Torabi Jahromi

Academic Rank: Assistant professor
Degree: Ph.D in Electrical Engineering
Phone: 09171023389
Faculty: Faculty of Intelligent Systems and Data Science


Development of dynamic fuzzy convolutional neural network structure and its application in hyperspectral image classification
Type Thesis
شبكه عصبي كانولوشن، شبكه فازي عصبي پويا، استخراج ويژگي، شناسايي الگو، طبقه بندي، تصاوير ابرطيفي Convolution neural network, Dynamic fuzzy neural network, Feature extraction, Pattern recognition, Classification, Hyperspectral image
Researchers fereshteh sharifi (Student) , Ahmad Keshavarz (Primary advisor) , Amin Torabi Jahromi (Primary advisor) , Valiollah Ghaffari (Advisor)


Research problem Deep learning architectures such as convolutional neural networks have recently gained a lot of popularity in the real world. The main reason for this popularity is that it can automatically extract features so that there is no need to extract and select manual features. In fuzzy neural networks, which integrates fuzzy systems and neural network, however, in fuzzy systems, it is difficult to identify the structure of the working neural network. One of the main problems is how to determine the number of hidden layers in such a way that an optimal and compact structure with high efficiency can be achieved. For this purpose, DFNN based on RBF neural networks, which are functionally equivalent (TSK), is used as a classifier. As mentioned in the introduction, due to the fact that DFNNs automatically add or remove neurons according to their function in the system and can adjust the structure and parameters at the same time, it is used as a classifier. In this research, we try to create an optimal structure with high performance for image classification by combining convolutional neural network and dynamic neural fuzzy. The importance and necessity of research In convolutional neural networks, fully connected layers are used for classification and due to the fact that there is a large number of neurons in fully connected layers, there is a convolutional neural network, and because of that, the number of network parameters increases. It is weighted and biased, which has a high computational volume. For this purpose, instead of the fully connected layers in the convolutional neural network, we will use the dynamic fuzzy neural network, which dynamically uses or removes the neurons according to their importance in the system's performance. We will be able to achieve image classification with less parameters and as a result less computational volume.