Implementation of A Convolutional Neural Network in A Website-Based Freshwater Fish Species Detection Application

Abstract

Indonesia has a high diversity of freshwater fish, but their identification and classification are often hindered by visual similarities between species. This study aims to develop a freshwater fish classification model using the Convolutional Neural Network (CNN) method with the MobileNetV2 architecture and implement it into a web-based application. The dataset consists of 1,565 images, covering 11 freshwater fish species and 1 non-freshwater fish class. The application was developed using the Extreme Programming (XP) methodology, including planning, design, implementation, and testing stages. Model evaluation achieved an accuracy of 87.12%, while functional testing of the web application confirmed that all features operated as intended. User testing with a Likert scale yielded an average satisfaction level of 92%. This study demonstrates that combining CNN MobileNetV2 with a Streamlit-based web application can provide a practical, fast, and accurate solution for freshwater fish detection.