MobileNets-V1 Architecture for Web Based Fish Image Classification

Abstract

Recently, the research study about fish identification become a very challenging to researchers. Climate and environmental changes have a major impact on fish species and their environment. To identify fish using manual process is time consuming and need effort to gather samples in different environment. The identification of fish species is performed by using feature extraction and a series of features. Generally, the characteristic is divided into two groups namely general characteristics and anatomical features. General characteristics is characteristic that can be seen directly without the aid of tools. The characteristics include color, texture, and fiber direction. Although, manual is performed by expert but is possible that identification is not accurate. Therefore, to overcome the problem, we create a web-based application for identifying fish by using image as input. We use 10 class data with 300 images for each class. Then, we split into training and testing with 80:20 ratio. The application was developed by using the MobileNets- V1 model. The proposed method has accuracy on 89 %, that obtain from training score is 91.04%, validation is 88,96%. This score is higher than other methods that used in this application. Total time for computation process is about 127 minutes.