Abstract—Abstract—The superior performances of convolutional neural networks in various fields have recently enabled deep learning to be popular. One of the most common problems in this regard is the determination of the structure of the deep artificial neural network according to the problem to be solved. In this paper, deep convolutional neural networks having different numbers of convolutional layers and different filter sizes are used for classifying challenging fish dataset. The results show that all of the tested structures succeeded in the learning data set, while the less deep structures with larger filters gave better results on the test data set. Increasing size of the filters may provide a performance boost up to 40.73 percent. In addition, tests were done by increasing the number of filters on each convolutional layer of successful structures. This operation led an extra performance boost up to 14.28 percent over the current performance of the structures.
Index Terms—Index Terms—Fish classification, deep learning, convolutional neural network, image classification.
M. Sarıgül is with the Computer Engineering Department, Cukurova University Sarıçam/Adana, Turkey (e-mail: msarigul@cu.edu.tr) & Computer Engineering Department, Iskenderun Technical University, İskenderun/Hatay, Turkey (e-mail: mehmetsarigul@iste.edu.tr). M. Avci is with the Biomedical Engineering Department, Cukurova University Sarıçam/Adana, Turkey (e-mail: mavci@cu.edu.tr).
[PDF]
Cite:M. Sarigül and M. Avci, "Comparison of Different Deep Structures for Fish Classification," International Journal of Computer Theory and Engineering vol. 9, no.5, pp. 362-366, 2017.
