SBIR Phase II: Advanced Computer Vision Methods for Diagnostic Medical Entomology

这项小企业创新研究（SBIR）第二阶段项目的更广泛影响是能够向国内和国际公共卫生机构提供高质量的病媒监测数据。病媒或将疾病传播给其他生物体的生物体，如蚊子和蜱虫，对人类健康和农业产生重大影响，具有相关的死亡率和发病率。该项目旨在推进人工智能方法，从高分辨率图像中识别蚊子种类。尽管得到了充分的研究和记录，但蚊子种类鉴定仍然是一项高技能的任务，在特定地区，少数能够掌握这项技能的人往往有许多其他工作职责，这使得将时间投入到蚊子鉴定的艰巨任务上很难证明是合理的，尽管创建数据是必要的。该项目及其衍生作品将使内部没有这种技能的组织能够获得这些非常有价值的数据。该解决方案还将允许拥有这种内部技能的组织将识别任务转移给季节性技术人员，并提供更大的数据集。这个更大的数据集将有助于更好地制定控制蚊媒疾病的决策。如果取得成功，这些方法可以转化为其他疾病媒介，进一步有利于公共卫生。这个小企业创新研究（SBIR）二期项目的核心是蚊子种类鉴定问题。世界上有3000多种蚊子，每种蚊子都有不同的行为和传播疾病的能力。经过地区训练的分类专家可以通过目视检查来识别它们，但这样的专家还很缺乏。一些基于图像识别的人工智能（AI）方法已经开发出来，但它们仅针对有限数量的物种设计，并且由于蚊子形态复杂以及病媒控制组织在实际使用中产生的可变性而面临问题。该项目旨在通过使用生成模型来解决由抽样偏差引起的训练数据集中的问题，从而增强基于人工智能（AI）的昆虫识别的现有方法。这些模型将用于调节未充分表示的属性的存在，以使模型更健壮，更少偏差。用于此任务的生成模型还将用于将数据在一个约束图像域转换为另一个约束图像域的可行性。最后的任务是使用这些模型来调整密切相关的蚊子物种的训练数据集，以微调细微但重要的区别。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 II project is to enable the provision of high quality vector surveillance data to public health institutions domestically and internationally. Vectors, or organisms that transmit diseases to other organisms, like mosquitoes and ticks, have a significant impact on human health and agriculture, with associated mortality and morbidity. This project aims to advance artificial intelligence methods to identify mosquito species from high resolution images. While well studied and documented, mosquito species identification remains a highly skilled task, where the few capable of this skill for a given region often have many other job responsibilities, making time devoted to the laborious task of mosquito identification difficult to justify at scale, despite the necessity of the data created. This project and its derivative works will enable organizations without this skill in-house to acquire this highly valuable data. The solution will also allow organizations with this skill in-house to task shift identification to seasonal technicians, and field a larger dataset. This larger dataset would enable better decision making for the control of mosquito borne disease. If successful, these methodologies can be translated to other vectors for disease, further benefiting public health.This Small Business Innovation Research (SBIR) Phase II project is centered around the problem of mosquito species identification. There are more than 3,000 species of mosquitoes in the world, each with different behaviors and capacities for carrying disease. Regionally trained taxonomic experts can identify them through visual inspection, but there is a shortage of such experts. Some artificial intelligence (AI) methods for image-based identification have already been developed, but they are only designed for a limited number of species and face issues due to complex mosquito morphology and the variability incurred in practical use by vector control organizations. This project seeks to enhance existing methodologies for artificial intelligence (AI)-based insect identification by making use of generative models to address issues in training datasets caused by sampling biases. These models will be used to modulate the presence of underrepresented attributes to make a more robust and less biased model. The generative models used for this task will also be used to translate the data for viability in one constrained image domain to another. The final task is to use these models to modulate the training datasets for closely related mosquito species to fine tune performance for minute, but important, distinctions.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.