뉴럴 네트워크를 이용한 수중 이미지에서의 실시간 물고기 인식 방법

Abstract

CNN composed of 24 convolutional layers and two fully-connected layers. We then trained the CNN with custom fish images. Actual fish images and videos are used to train and evaluate the CNN. The CNN recorded 93% detection accuracy and ran at 16.7 frames per second (FPS). The proposed method can detect fish in dim, noisy, and hazy underwater optical images precisely and accurately in a real time without preprocessing of images