Development of an accurate automated fish weight and distribution camera for fish farms

为养鱼场开发精确的自动化鱼类重量和分布相机

• 批准号: 10038840
• 金额: $ 44.59万
• 依托单位: ACE AQUATEC LIMITED
• 依托单位国家: 英国
• 项目类别: Collaborative R&D
• 财政年份: 2022
• 资助国家: 英国
• 起止时间: 2022 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=10038840
• 关键词: Development; accurate; automated; fish; weight

Fish farms waste feed (60% of production costs, 'Mowi Handbook 2022) by over feeding dominant fish, and underfeeding smaller fish (which leads to stress, higher incidence of disease and parasites, such as sea lice). Farmers attempt to rectify this through manual random sampling (30-40 fish per pen, on 12 cage sites) to provide indications of low/average/high fish weights. There are around 400 live sites in Scotland each year with typically 12 cages per site containing 15-17.5kg of fish per cubic tonnes of seawater (1.5/98.5 water to fish ratio). Stocking density is heavily regulated so accurate biomass data is critical for compliance. Typically a farm will grade out smaller fish into a new pen to bring all fish up to the required harvest weight. However, inaccurate data (due to sampling) leads to many fish being harvested above or below the required weight. Supermarkets do not compensate for under or over production. Wasted 'consumed' feed (at £1300 per tonne) costs farmers a significant proportion of their income (every kilo over the required weight costs £4.43, when the average cost of a fish is bought for £40 (Kontali 2019)). Tools such as Vaki's Biomass Daily frame, Akva's Vikas, or several new AI cameras (Aquabyte, Optoscale) have attempted to automate the weight estimation process; but limited visibility throughout the pen leads to misleading distribution data. While farmers have better tools to track average growth in the pen ensuring compliance, they have little help when accurately assessing when to grade fish to prevent feed waste. Ace's innovation utilises multiple stereo cameras embedded in a 360 arrangement to provide the first 'depth' point of cloud mapping and AI weight estimation, capable of tracking fish in all parts of the cage providing high fidelity distribution accuracy, and a clear portal for informing fish husbandry and saving money.

养鱼场浪费饲料（占生产成本的60%，'Mowi Handbook 2022），过度喂养优势鱼类，并喂养较小的鱼类（导致压力，疾病和寄生虫的发病率较高，如海虱）。养殖户试图通过人工随机抽样（12个网箱，每个围栏30-40条鱼）来纠正这一点，以提供低/平均/高鱼重的指标。苏格兰每年约有400个活鱼地点，每个地点通常有12个网箱，每立方吨海水中含有15- 17.5公斤的鱼（水鱼比为1.5/98.5）。放养密度受到严格管制，因此准确的生物量数据对于合规至关重要。通常，养殖场会将较小的鱼分级放入新的围栏，以使所有鱼达到所需的收获重量。然而，不准确的数据（由于采样）导致许多鱼被捕捞超过或低于所需的重量。超级市场不能弥补生产不足或过剩。浪费的“消费”饲料（每吨1300英镑）花费了农民收入的很大一部分（每公斤超过所需重量的成本为4.43英镑，而购买鱼的平均成本为40英镑（Kontali 2019））。Vaki的Biomass Daily框架、Akva的Vikas或几种新的AI相机（Aquabyte、Optoscale）等工具试图自动化重量估计过程;但整个围栏的可见性有限导致误导性的分布数据。虽然农民有更好的工具来跟踪围栏中的平均生长，以确保合规性，但在准确评估何时对鱼类进行分级以防止饲料浪费时，他们几乎没有帮助。Ace的创新利用嵌入360度布局的多个立体摄像头来提供云映射和AI重量估计的第一个“深度”点，能够跟踪笼子所有部分的鱼，提供高保真的分布准确性，以及一个清晰的门户，用于通知鱼类饲养和节省资金。