Acquacoltura Pubblicazione scientifica In Situ Sea Cucumber Detection Based on Deep Learning Approach FILONE TECNOLOGICO TEMA economia circolare e sostenibilità produzione e raccolta RIASSUNTO Aquaculture provides abundant food that directly consumed by humans. However, aquaculture cultivation is mainly replies on manual management. In order to improve the production efficiency and reduce the labor costs, numbers of recently developed techniques are applied to promote the automation level in aquaculture. In this work, we initiate the study of in situ marine organisms monitoring. Sea cucumber was chosen as the detection target in this paper because the cultivation of sea cucumber has grown quickly in Asian countries. A sea cucumber detection scheme was implemented by utilizing the recently developed YOLO v2 model. Sea cucumber detection model was trained on images selected from one video clip. Experiments were conducted on the other frames from the same video clip. The detection accuracy of 97.6% was achieved. In addition, sea cucumber detection was tested on images collected from the internet. The detection accuracy on new data was 76.3%. The detection ability on new targets was remarkable by the deep learning approach. Individual sea cucumbers presenting in various poses or partially occluded were accurately detected in natural scene. DATA Data di pubblicazione: 01/05/2018 AUTORI CHUNLEI XIA LONGWEN FU HUI LIU LINGXIN CHEN ENTE DI AFFERENZA YANTAI INST COASTAL ZONE RSCH CHINESE ACAD SCIENCES YANTAI CHINA RIVISTA 2018 OCEANS - MTS/IEEE Kobe Techno-Oceans (OTO) (Page(s): 1-4)