A Smart Aquaculture Feeding System Using Deep-Learning Based Underwater Image Object Recognition

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

MOST109-2221-E019-059

Translated Name/計畫中文名

基於深度學習水下影像物件識別技術之智慧投餌系統

Project Coordinator/計畫主持人

Shyi-Chyi Cheng

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=13540744

Year

2020

Start date/計畫起

01-08-2020

Expected Completion/計畫迄

31-07-2021

Bugetid/研究經費

812千元

ResearchField/研究領域

資訊科學--軟體

本計畫結合深度學習開發可應用於智慧投餌之水上、水下影像辨識關鍵技術，空拍無人機拍攝水面影像，用以建立水上3D模型，自動收集養殖場域資料、監控環境及偵蒐可疑物件；本計畫擬發展一自動水面無人船導航技術，配置水下立體攝影機，監控水下魚群及設施，藉以辨識養殖魚群生長狀況。首先，固定式水上水下攝影機聲納影像用以建立水下3D物件偵測與辨識技術，以利無法使用無人載具時，系統仍能持續監控養殖場域，提供全天候之完整場域監控功能。本計畫可概分為三部分:(1)主要針對融合水下立體攝影機或聲納，建立3D影像辨識之深度學習模型；(2)使用空拍影像定位場域物件位置，指揮無人船建立移動式水下立體影像，進行水下魚隻、魚群行為辨識；(3)開發可用優化養殖模式的智能化投餌系統。本計畫之總計畫以智慧養殖的主要應用領域，發展相關的關鍵技術，包括水下立體影像魚群偵測、水下聲納影像魚隻偵測、魚體辨識、養殖場域監控分析與水面可疑物件分析。從魚隻偵測的結果，藉由游動軌跡分析，可以監控魚群生長狀態，例如生病魚隻發生事件；魚群活力分析另一個重要的應用是提供即時投餌資訊，與自動投餌機結合，我們可以進一步發展魚群搶時活力辨識技術，確認投餌最佳化，透過空拍機或無人船連動投餌裝置，降低養殖成本。發展智慧養殖所需的辨識技術，並與台灣海洋大學養殖學系、農委會漁業署合作，合作單位提供實驗場域與相關設備，一起發展具養殖專業知識之養殖場域事件分析與預測系統，所建立之資料庫，結合大數據分析技術，可以最佳化魚群養殖模式。 This project presents deep-learning based underwater image processing and recognition for smart aquaculture feeding. The system uses an autonomous drone to capture image sequences for the target aquaculture farm modeling, data collection, monitoring, and event prediction. An autonomous ship, equipped with an underwater stereo camera is also used to monitor underwater objects for inspecting the growth and behaviors of fish school. Finally, to achieve the goal of intelligent fish feeding, a cloud system is constructed to execute the deep-learning models and store the collected fish-related data into the database. The deep-learning models for detecting and recognizing 3D objects in underwater stereo and sonar images are first trained based on the collected dataset captured by the fixed cameras which are still necessary because autonomous drone or ship is often out of services in bad weather days. This project is divided into three parts: (1) the deep-learning models for underwater 3D object detection and recognition are trained using fixed cameras; (2) both autonomous drone and ship are used to capture underwater images for understanding the behaviors of fish and fish school; (3) a smart fish feeding system is finally designed based on the techniques developed in the first and second parts. The project is a sub-project of the integration one which focuses on the development of underwater environment monitoring for smart aquaculture farming. The techniques under development include image recognition, water-quality inspecting, and intelligent robots for aquaculture farm monitoring. The project co-operates with Department of Aquaculture at National Taiwan Ocean university and Fisheries Agency, Council of Agriculture, Taiwan. The co-operative departments would provide us the experimental equipment and environments for building up meaningful aquaculture database which is important in developing new smart aquaculture artificial intelligent services.

Keyword(s)

智慧投餌
深度學習
水下3D物件偵測
聲納影像
立體影像
魚群搶食強度辨識
無人機
無人船
intelligent fish feeding
deep learning
underwater 3D object recognition
sonar image processing
fish feeding intensity recognition
drone
autonomous ship

DSpace CRIS

A Smart Aquaculture Feeding System Using Deep-Learning Based Underwater Image Object Recognition

Details

Description