SBIR Phase I: Advanced‌ ‌Computer‌ Vision‌ ‌Methods‌ ‌for‌ ‌Mosquito‌ ‌Surveillance

这项小企业创新研究（SBIR）第一阶段项目的更广泛影响将来自开发防止未来蚊子疾病爆发的新工具。大多数美国蚊子控制组织缺乏进行常规蚊子监测的能力或能力，这是有效控制蚊子的必要任务。在全球范围内，80%的世界人口面临蚊媒疾病的风险，蚊虫监测、监测和评价被广泛认为是需要在全球推广的重要公共卫生活动。通过该提案开发的技术将开发并证明新的鉴定方法的可行性，以降低蚊虫监测的操作成本，同时提高数据的准确性和标准化。其结果将是改进决策，减少蚊媒疾病的发病率。这个小型企业创新研究（SBIR）第一阶段项目将建立在计算机视觉的进步基础上，以便在操作环境中高精度地识别蚊子种类。虽然使用深度卷积神经网络（cnn）已经证明了蚊子物种的高精度分类，但证据仅限于受控的实验室环境，少数物种的小型数据集或实验室饲养的标本。操作环境面临着一个复杂得多的问题，可能遇到数百种潜在的物种，以及野生捕获的蚊子的形态和质量的变化。本提案旨在克服和减轻当前研究状态未解决的核心技术挑战，包括：精细分类技术需要从自然界中发现的3000多种具有明显重叠形态的蚊子中区分医学相关的物种，新的物种检测方法可以识别所呈现的标本是否来自物种分类算法未知的物种，以及表征标本的取食状态和物理质量，如翅膀、腿、鳞片和身体的损伤。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will result from the development of new tools to prevent future mosquito disease outbreaks. Most US mosquito control organizations lack the capability or capacity to conduct routine mosquito surveillance, a necessary task for effective mosquito control. Globally, 80% of the world population is at risk for mosquito-borne disease, and mosquito surveillance, monitoring, and evaluation are widely recognized as critical public health activities o be scaled globally. The technology developed through this proposal will develop and demonstrate the feasibility of new identification methods to reduce operational mosquito surveillance costs, while improving accuracy and standardization of data. The result will be improved decisions that will reduce the incidence of mosquito-borne diseases. This Small Business Innovation Research (SBIR) Phase I project will build on advances in computer vision for high accuracy identification of mosquito species in operational contexts. While high accuracy classification of mosquito species has been demonstrated using deep convolutional neural networks (CNNs), evidence has been limited to controlled laboratory environments, with small datasets of few species, or with lab reared specimens. Operational environments face a significantly more complex problem, with hundreds of potential species that may be encountered, and variation in morphology and quality of wild-caught mosquitoes. This proposal seeks to overcome and mitigate the core technical challenges unaddressed by the current state of research, including: fine-grain classification techniques required to distinguish medically relevant species from over three thousand mosquito species found in nature with significant overlapping morphology, novel species detection methods to identify when a presented specimen is from species unknown to the species classification algorithms, and characterizing the feeding state and the physical quality of specimens, such as damage to wings, legs, scales, and body.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.