Computer vision-based detection of fish from acoustic backscatter time series

我们提出的研究解决了使用计算机视觉技术从回声探测仪数据中检测视觉模式的新方法。从更传统的声学数据处理模式切换到**计算机视觉算法可能会为回声探测仪数据提供新的见解，并可能使我们的合作伙伴**公司能够丰富他们的自动数据处理服务组合。**我们的工业合作伙伴是ASL环境科学公司(ASL)，这是一家总部位于不列颠哥伦比亚州**的员工所有的公司。ASL是环境监测创新解决方案的全球领导者，在海洋、声学、遥感和冰川研究产品以及科学**咨询服务方面拥有超过40**年的经验。自2000年以来，ASL一直在建造声学背向散射回声探测仪。到目前为止，ASL为处理从其声学反向散射测深仪收集的数据提供的支持**非常有限。**考虑到ASL制造的165台回声测深仪，客户向**非声学家的转变，以及该技术在更多应用中的采用，现在有**机会为ASL的产品组合增加新的数据处理服务。ASL有兴趣探索一些潜在的应用**领域，例如用于对悬浮沉积物、碎石冰、海洋湍流、水中油、浮游动物/虾和鳍鱼进行分类的自动数据处理。**这项合作将使有重点的短期(六个月)可行性研究能够从计算机视觉的角度解决**鳍鱼检测问题。我们将利用超声图像的视觉线索**利用计算机视觉算法对鱼进行检测和分类。本课程将探讨传统的分词方法，以及深度学习的最新发展。

Our proposed research addresses novel ways of detecting visual patterns from echosounder data using**computer vision techniques. Switching from the more traditional acoustic data processing paradigm to**computer vision algorithms may provide new insights into echosounder data, and may enable our partner**company to enrich their portfolio of automatic data processing services.**Our industrial partner is ASL Environmental Sciences Inc (ASL), an employee-owned company based in**Victoria, BC. ASL is a global leader in innovative solutions for environmental monitoring and has over 40**years of experience in oceanographic, acoustic, remote sensing and ice research products and scientific**consulting services. Since 2000, ASL has been building acoustic backscatter echosounders. To date, ASL has**provided very limited support for processing the data gathered from its acoustic backscatter echosounders.**Considering the 165 echosounder units that ASL has manufactured, the change in the clientele towards**non-acousticians, and the adoption of the technology for a greater number of applications, there is now an**opportunity to add new data processing services to ASL's portfolio. There are a number of potential application**areas that ASL is interested in exploring, such as automatic data processing for classification of suspended**sediments, frazil ice, ocean turbulence, oil in water, zooplankton/shrimp and fin fish.**This Engage collaboration will enable a focused short-term (six months) feasibility study addressing the**problem of fin fish detection from a computer vision perspective. We will exploit visual cues from echograms**to detect and classify fish with computer vision algorithms. Traditional segmentation methods, as well as recent**developments in deep learning will be explored.