I-Corps: Identification of disease-carrying mosquito vector species using computer vision

这一I-Corps项目更广泛的影响/商业潜力是开发蚊子媒介监测方法，使公共卫生机构能够更好地评估媒介传播疾病爆发的风险。对蚊子等病媒种群的监测、监测和评估被广泛认为是一项战略性公共卫生活动，需要在全球范围内推广。然而，目前的做法需要大量的后勤资源和昆虫学专业知识，在美国和世界各地的许多地方都无法获得。在美国，近一半的蚊子控制组织缺乏进行常规监测的能力。这是复杂的不同区域的做法，地理，资源和当地蚊子物种组成。计算机视觉是一种技术方法，可以提供标准化的能力来分类不同种类的蚊子。该技术可以嵌入到实验室、现场和远程部署设备的各种配置中，降低监控成本，同时为决策提供更准确的数据。这项技术的灵活性可能允许它被修改为其他application.This I-Corps项目是基于一个动态的光学配置和计算机视觉系统的发展，能够高精度的蚊子物种分类。使用深度卷积神经网络（CNN），该平台成功地实现了94.7%的分类准确率，使用计算机视觉对24种野外收集（高度受损）的标本进行了分类，而昆虫学家的准确率为66%。光学系统优化了景深、颜色、照明和分类所需的分辨率。为了提高蚊子控制项目进行监测的能力，并提高数据质量，以便更有效地做出决策，该技术包括用于远程蚊子分类的自动捕蚊器，该奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响进行评估，被认为值得支持审查标准。

The broader impact/commercial potential of this I-Corps project is the development of mosquito vector surveillance methods that enable public health agencies to better evaluate risk of vector-borne disease outbreaks. Surveillance, monitoring, and evaluation of vector populations, such as mosquitoes, is widely recognized as a strategic public health activity that will need to be scaled globally. However, current practices require significant logistical resources and entomological expertise unavailable in many parts of the US and around the world. In the US, nearly half of mosquito control organizations lack the capability and capacity to conduct routine surveillance. This is complicated by varying regional practices, geographies, resources, and local mosquito species compositions. Computer vision is a technical approach that may offer standardized ability to classify different species of mosquitoes. The proposed technology could be embedded in various configurations from lab, field, and remotely deployable devices, reducing surveillance costs while providing data with greater accuracy for decision-making. The flexible nature of this technology may allow it to be modified for additional applications.This I-Corps project is based on the development of a dynamic optics configuration and computer vision system capable of high accuracy classification of mosquito species. Using deep convolutional neural networks (CNN), the platform successfully achieved 94.7% classification accuracy across 24 species of field collected (highly damaged) specimens using computer vision compared with 66% accuracy among entomologists. The optical system optimizes depth of field, color, lighting, and resolution necessary for classification. To increase capacity for mosquito control programs to conduct surveillance, and improve data quality for more effective decision-making, the technology includes automated mosquito traps for remote mosquito classification, a robotic identification and sorting system and a handheld identification and sorting tool.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.