A Fish Identification System Based On Deep Learning with Convolutional Neural Network

• Main Authors: Hsieh, Hsin-Yin, 謝昕穎
• Other Authors: Tseng, Ching-Hsiang
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/3mzag8

碩士 === 國立臺灣海洋大學 === 電機工程學系 === 107 === Currently, all information about the fishes in the National Museum of Marine Biology &Aquarium (NMMBA) in Taiwan is provided through on-site explanation boards or audio tours. However, with so many different fish species in a large fish tank, the explanation board has to be enormous and takes up a lot of space, and an audio tour could also be very manpower demanding. A potential solution to this problem is to develop a fish identification system which can identify fish species quickly and be operated easily. The focus of this research is to build a fish species identification system which links to a fish information database. Users can take a picture of the fish with their mobile phones and upload the fish image to the system. The system would identify the exact fish species quickly and extract information about the fish from the database, so that the users can gain knowledge about the fish immediately. The development of the fish identification system is base on deep learning with Convolutional Neural Network. This research uses the MATLAB as the software platform and different kinds of aquarium fish as samples to achieve a correct identification rate of 90%, which is a quite promising result. We will keep increasing identifiable fish species and hope to include every common aquarium fish in Taiwan.