Abstract
細部計畫1:臺灣周邊海域箱網抗風浪研究試驗＜/strong＞＜/P＞＜P＞本計畫旨在研究台灣周邊海域箱網的抗風浪能力，藉由實測分析波浪、海流、錨碇張力和網袋變形之間的關聯，深入探討海上箱網的力學特性，提供政府及箱網養殖業參考。本計畫將在颱風季節於箱網養殖區設置ADCP波流儀，以量測波浪及海流；在上流面的纜繩上安裝張力計，以量測錨碇張力；並在浮框和沉框安裝水位計，以分析網袋變形。此外，本計畫將在東北季風期間，在箱網的陸側安裝張力計，實測分析落山風對於箱網的影響。最後，本計畫將利用自行研發的數值模式，分析錨碇點沉陷或滑動，對於箱網力學特性的影響。＜/P＞＜P＞＜br /＞＜strong＞細部計畫2:箱網養殖主要及新興物種研究＜/strong＞＜/P＞＜P＞針對箱網養殖潛力物種紅九棘鱸種魚透過分子標記及遺傳經濟性狀轉錄體建立基因資料庫與追蹤子代親緣關係，透過生物分子育種技術進行魚種之種魚親源鑑定與基因多樣性分析工作建立種魚族群個體標幟與記錄追蹤系統、基因資料庫及微衛星DNA標記等遺傳資訊，以供後續可利用分子標記來輔助評估紅九棘鱸種原庫遺傳多型性與篩選具重要經濟性狀之候選種魚等資訊，再從具有差異性體表型成績的魚隻中，藉由次世代轉錄體定序所建立之功能性基因庫並篩選出具有與重要經濟性狀遺傳連鎖效應之簡單序列重覆（simple sequence repeats，SSRs）或單一核苷酸多型性（single nucleotide polymorphisms，SNPs）等分子標記，另針對箱網養殖物種進行疾病調查建立養殖疾病管理與防治。探討流行病監測及虹彩病毒攻擊感染模式，以了解病毒感染和寄主免疫反應等相關研究；並開發紅九棘鱸病毒及細菌性疾病之不活化疫苗，有效預防疾病所造成之損害。同時持續透過養殖管理建立孵化養殖技術及高生物安全飼養系統稚魚模式，且透過蒐集箱網養殖物種生產管理及成長資訊，整合智慧化箱網養殖 AI 運算數據資料庫應用並進行競爭力分析。最後透過準確的檢測技術監控和追踪，了解乳酸桿菌分佈情況和動物的胃腸道中的數量，並分析生物學特性有效使用和益生菌的開發。＜/P＞＜P＞＜br /＞＜strong＞細部計畫3:智能化箱網養殖模式研究＜/strong＞＜/P＞＜P＞ 人工智慧物聯網(AIoT)技術日益成熟，擴充現有的水產養殖設備的資通訊(ICT)連結能力及養殖場域的監控裝置智慧化，本計畫以AIoT系統收集高品質養殖數位庫邁出養殖技術數位化的第一步，進而以機器學習/深度學習及資料探勘挖掘隱藏在養殖數據庫背後的專家養殖規則，用來進行自動控制養殖裝置貫徹專家養殖模式，進而達成節省成本、提升產量等智慧養殖目的。配合企業資源規劃系統( ERP)管理產銷供應鏈及先進規劃排成系統(APS)進行精確的養殖規劃，所建置的智慧養殖AIoT系統得以輔助建立以「智慧養殖」及「數位服務」為宗旨的新型態智慧養殖服務產業。＜/P＞＜P＞ 本計畫目標為定義一可複製、整廠輸出的AIoT智慧養殖系統組態，並據以建立優質的智慧養殖數位服務。智慧養殖的主要目的不外乎在減少人力負擔、即時監控魚群健康及行為、精確執行養殖模式，以提升產量、完善養殖規畫。以精準餵食模型為例，養殖專家依養殖池之魚隻平均體重、魚隻數量、魚種校正因子、水溫、水流速等決策因子計算飼料投餌量。本計畫預期透過機器學習之回歸模型訓練演算法，建立投餌量預測模型，用來設定及控制智能投餌機，完成精準餵食的目標。除此之外，以需求訪談明確定義本AIoT系統，得以建立有意義的養殖決策模型，完備精準智慧養殖所需的數位服務。這樣的系統可減少人工因情緒、體能及其他非理性的因素造成的精確執行的偏差。透過雲端智慧養殖管理平台進行養殖池魚群資訊數位記錄與資料儲存，藉由已收集的養殖數據庫，進行資料探勘，建立適切的精準養殖決策模型。養殖決策模型制定之養殖模式必須隨時間或場域調適，持續觀察魚群行為變化，依機器學習之強化學習技術，可以用來制定動態養殖決策模型。＜/P＞＜P＞ 養殖設備(如投餌機)配備嵌入式IoT傳感器，以進行魚群成長監測和動態行為記錄，可將施錯餌料的機率降低80％；同時養殖機具利用AI和機器學習來預測設備故障時程，進而提前安排維護工作，縮短維修時間、提升智慧養殖設備的妥善率及可靠度，賦予智能化、雲端鏈結能力，才能確保減少人力負擔的目標得以達成。採用攝影機架設於分魚機水道進行魚隻分級機構的設計，可準確測量箱網中的魚隻體長，並計算箱網內部魚隻數量，結合雲端大數據分析平台，進一步優化分魚機的機構設立，達到精準的養殖規劃。＜/P＞＜P＞ 本計畫包含六項研究子計畫：＜/P＞＜P＞(1) AIoT為基礎的養殖決策系統架構分析。＜/P＞＜P＞(2) AIOT為基礎的養殖決策系統建置。＜/P＞＜P＞(3) 低成本、自主技術智能投餌機設計。＜/P＞＜P＞(4) 具非侵入式魚隻資訊測量及魚隻大小分類的分魚機設計。＜/P＞＜P＞(5) 具自動雲端連線能力的智慧型自主式即時聲納浮型載具。＜/P＞＜P＞(6) 具自動雲端連線能力的智慧型自主式即時攝影浮型載具。＜/P＞＜P＞ 每項子計畫均包含外掛在共用之雲端平台上的相對應AI服務，得以完善本計畫成果的整廠輸出經營模式。本細部計畫統合各子計畫之研究，進行各子計畫之資料庫建立，並將成果整合成一具設備自動化能力之「雲端化AIoT智慧養殖」平台，建置可提供智慧養殖數位服務的AIoT系統，並於實際場域進行實際成果的管理及服務測試。This project aims to study the anti-typhoon ability of net cages in the open water around Taiwan. In this study, the relationship among waves, ocean currents, mooring tensions, and net-cage deformations will be figured out via field measurements to provide further information on net cages for the government and the industry. In this project, one ADCP (wave and current meters) will be installed in the cage culture area during typhoon season to measure waves and ocean currents; one load cell will be installed on the mooring line on the upstream side to measure mooring tensions, and a set of water level loggers will be installed on the floating collar and the tube sinker to evaluate the deformation of the net cage. In addition, this project will set a load cell on the land-side mooring line of the net cage during the winter monsoon period to measure and analyze the impact of the downhill wind on the net cage. Finally, this project will use the self-developed numerical model to simulate the impact of anchor point settlement or slippage on the mechanical characteristics of the net cage.＜/P＞<br/><b>2:Studies on the main and emerging species in cage culture</b>＜P＞To establish a gene database and track the genetic relationship of offspring based on the potential species of box net culture, Red Ninefish, through molecular markers and transcripts of genetic economic traits, use biomolecular breeding technology to identify the source of fish species and analyze the genetic diversity of the fish species. Establish genetic information such as individual flagging and record tracking system, gene database, and microsatellite DNA markers for species fish populations, so that molecular markers can be used to assist in the subsequent evaluation of the genetic polymorphism of the original bank of red nine perch species and the selection of important economic traits Then, from the fish with different body phenotypic performance, the functional gene library established by next-generation transcript sequencing is used to screen out simple sequence sequences that have genetic linkage effects with important economic traits. Molecular markers such as simple sequence repeats (SSRs) or single nucleotide polymorphisms (SNPs), and also conduct disease investigations for box net species to establish breeding disease management and prevention. Discuss epidemic surveillance and iridescent virus attack infection mode to understand virus infection and host immune response and other related research; and develop inactivated vaccines for red nine perch virus and bacterial diseases to effectively prevent damage caused by diseases. At the same time, it continues to establish hatchery breeding technology and high biosafety breeding system juvenile fish model through breeding management, and integrates the application of intelligent cage net breeding AI computing data database and conducts competitiveness analysis by collecting production management and growth information of cage net breeding species. Finally, through accurate detection technology monitoring and tracking, understand the distribution of lactobacilli and the number of animals in the gastrointestinal tract, and analyze the effective use of biological characteristics and the development of probiotics.＜/P＞<br/><b>3:A Study of Artificial Intelligence on Smart Cage Culture</b>＜P＞ With the rapid advance of AIoT technologies, in this project, we first design an AIoT- based smart aquaculture system by adding wireless communication ability into current existing aquaculture machines and surveillance devices which automatically collect high-quality data from aquaculture ponds and offshore cages. The collected big dataset digitizes the underlying aquaculture expertise and guides the development of the smart aquaculture rules by training certain fishing feeding prediction models using machine learning, deep learning or data mining techniques. These models are then implemented in our smart aquaculture cloud, which likely operates as the brain of an aquaculture expert to execute precise aquaculture rules based on the aquaculture machines powered by the AIoT technologies. Our aquaculture cloud offers smart digital services to the aquaculture farms to achieve the goals of smart aquaculture including cost reduction, labor savings, and increased profit. For an aquaculture farm, the AIot smart aquaculture system is the manufacturing executing system (MES) which can further cooperate with the enterprise resource planning (ERP) and advanced planning scheduling (APS) systems to offer smart aquaculture techniques and digital services to aquaculture farms with the intention of achieving the goal of digital transformation.＜br /＞ This project focuses on the definition of a repeatable and transferable business model that guides the transformation of the developed AIoT smart aquaculture system. To achieve the goal of increased production and excellent aquaculture planning, the decision-making models for labor-saving, real-time inspection of the fish health and behaviors, and precise execution of aquaculture expertise should be implemented into the software agent. With the estimation of fish bait as an example, the expertise determines the amount of bait to feed a fish pond by multiplying the values of variables including the water temperature, the average fish weight, the number of fish, and a species-dependent correction factor. The cloud estimates the amount of the fish feeding bait with the pre-trained regression neural network based on the collected dataset. The amount of fish food is then added into our smart fish feeding machine to precisely execute the fish feeding tasks. Although we build up all the decision models based on the high-quality dataset, they should be carefully corrected by the user requirements which requires a series of communication with the workers of the experimental aquaculture farm in order to complete a practical AIoT system for smart aquaculture. The contributions of the system include (1) well-verified digital AI services are implemented on the cloud; (2) human bias in fish feeding is reduced; and (3) aquaculture expertise is digitized and stored in the smart aquaculture cloud which is equipped with important facilities to ensure the information security and increase the system reliability. Based on the digitized aquaculture expertise, i.e., the collected dataset, it is possible to transfer our AIoT system to another aquaculture farm by adapting the pre-trained decision models using the farm-specific data using machine learning techniques. ＜br /＞Furthermore, online monitoring and learning algorithms should be studied accordingly to capture dynamic aquaculture decision models. Farm equipment (such as feeders) equipped with embedded IoT sensors can monitor fish growth and capture fish dynamic behaviors. In this way, it can reduce the possibility of employing an incorrect baiting method by 80%. Additionally, the integration of AI and machine learning in farming equipment can predict the course of equipment failure time. Furthermore, with this prediction capability, maintenance work can be arranged in advance. Thus, it shortens maintenance time, improves the appropriate rate and reliability of smart breeding equipment, and provides added intelligence and cloud linking capabilities to ensure and achieve the goal of reducing labor and work requirements in managing aquaculture farms. The design of the fish grading mechanism by using the camera settings in the fish channel of the sorting machine can accurately measure the body length of the fish and calculate the number of fish in the aquaculture cage. Combined with the cloud platform's big data analytics, the fish sorting machine is further optimized and helps aquaculture farms to establish and achieve precise feeding planning.＜br /＞ This project serves to upgrade and improve our existing "Cloud-based AIoT Smart Farming" platform as it aims to develop appropriate digital AI services for smart farming. Based on the digitized aquaculture expertise, i.e., the collected dataset, the proposed AIoT system can be implanted into other aquaculture farms by transfer learning based on further data collections. It also explores the feasibility of the business model to transfer the whole AIoT system across aquaculture farms. The purpose of the research is the following.＜br /＞• Develop a digitization methodology for aquaculture expertise to offer digital AI services for precise aquaculture based on the "Cloud-Based AIoT Smart Aquaculture" platform.＜br /＞• Build an AIoT-based aquaculture decision-making system.＜br /＞• Design a low-cost smart fish feeding machine of self-owned development technique.＜br /＞• Design a fish sorter with non-invasive fish information measurement and fish size classification.＜br /＞• Construct a mobile underwater / aquatic image analysis platform.＜/P＞＜P＞This project contains six research sub-projects:＜br /＞(1) Analysis of the architecture of the AIoT-based aquaculture decision-making system.＜br /＞(2) Establishment of an AIoT-based feeding decision-making system.＜br /＞(3) Design of a low-cost intelligent fish feeding machine with a self-owned development technique.＜br /＞(4) Design of a fish sorter for non-invasive fish measurement and fish size classification.＜br /＞(5) Intelligent autonomous real-time sonar floating vehicle with automatic cloud connection capability.＜br /＞(6) Intelligent autonomous real-time camera floating vehicle with automatic cloud connection.＜br /＞ Each sub-project has a corresponding AI service plugged into the shared cloud platform, which can improve the whole-system output business model of the project results. This project integrates each sub-project, establishing their corresponding database component, together with the results, which are added to the "cloud-based AIoT smart aquaculture" platform with equipment automation capabilities. It also builds a platform that provides smart farming technology services using an AIoT system.＜/P＞＜P＞＜br /＞ ＜/P＞
