Abstract
The diversity of fish species in coral reef ecosystems is one of the indications in determining health in coral reef ecosystems. Many Indonesian Fisheries and Marine Research and Development Agency experts carefully classify fish images. A reliable technique for performing image classification is Convolutional Neural Network (CNN). Transfer learning appears and adopts part of CNN, namely the modified convolution layer. The paper aims to solve the fish classification problem using the pre-trained model of Mobilenet V2. The model has a low computational process and does not use too many memory resources when training image data. The research image data used is 49,281 data of various sizes and 18 types of fish. The image is entered into the transformation process (random rotation, random resize crop, random horizontal flip) on the training and test data to produce varied data. After the transformation process, the image data is entered into the training process using the Mobilenet V2 architecture. Testing the Mobilenet V2 architectural model obtained an accuracy score of 99.54%, which is reliable in classifying fish images.
DOI
10.17977/um018v5i12022p67-77
Recommended Citation
Suhana, Rizka; Mahmudy, Wayan Firdaus; and Budi, Agung Setia
(2022)
"Fish Image Classification using Transfer Learning Method with
Adaptive Learning Rate,"
Knowledge Engineering and Data Science: Vol. 5:
No.
1, Article 6.
DOI: 10.17977/um018v5i12022p67-77
Available at:
https://citeus.um.ac.id/keds/vol5/iss1/6