SBIR Phase I: Exploiting High-Resolution Imagery, Geospatial Data, and Online Sources to Automatically Identify Direct Marketing Leads

这个小企业创新研究第一阶段项目将进行可行性研究，以证明通过结合目前可用的高分辨率图像、地理空间数据（例如，包裹数据或结构数据）和其他相关的在线数据源（例如，财产税数据或人口普查数据），有可能自动为各种市场产生高度针对性的直接营销线索。计划的方法这个问题,(1)将现有的地理来源与高分辨率图像,以确定确切位置和确定包裹的地址的意象,(2)从图像中提取相关特征提供适当的引导,例如确定一个游泳池的存在与否,屋面材料的类型,或什么类型的汽车停在车道上,和(3)引进其他来源的数据,如物业税评估数据提供额外的背景。第一阶段项目的主要重点将是展示使用机器学习技术来识别可用于直接营销的高分辨率图像中的特征。高分辨率的航空图像现在正在广泛收集，并且可以以低成本获得，在某些情况下甚至是免费的。挑战首先是将包裹数据与高分辨率图像对齐，以识别属性的确切地址和边界，其次是开发特征提取技术，可以利用上下文信息准确识别新特征，如屋顶，汽车，游泳池，景观美化等，可用于直接营销。准确识别图像特征，然后将其与特定属性以及相关信息来源联系起来的能力，将使有针对性的直接营销产品得以建立。该产品的最终用户将是寻求直接向住宅消费者销售产品的公司。这包括与家庭装修相关的产品和服务，包括外部和内部，以及与家庭居民相关的产品，这些产品可以从有关包裹的可用图像中收集到。这是一个巨大的市场，包括从家庭装修商店到屋顶公司、建筑公司、汽车经销商、树木修剪商、园林设计师和游泳池建筑公司。除了直接营销，该技术还可用于结合图像、地理空间数据和结构化信息的其他应用。例如，它可以用于减少蚊子，这对阻止西尼罗河病毒的传播很重要，通过识别大的死水池，将这些危险与适当的地址联系起来，然后向居民邮寄减少通知。

This Small Business Innovation Research Phase I project will conduct a feasibility study to demonstrate that by combining currently available high-resolution imagery, geospatial data (e.g., parcel data or structure data), and other related online data sources (e.g., property tax data or census data), it is possible to automatically generate highly targeted direct marketing leads for a variety of markets. The plan is to approach this problem by (1) aligning existing geospatial sources with the high-resolution imagery in order to determine the exact location and determine the address of the parcels seen in the imagery, (2) extracting the relevant features from the imagery to provide appropriate leads, such as determining the presence or absence of a swimming pool, the type of roofing materials used, or what types of cars are parked in the driveway, and (3) bringing in other sources of data, such as property tax assessment data to provide additional context. The primary focus of the phase I project will be to demonstrate the use of machine learning technology for identifying features in high-resolution imagery that can be used for direct marketing. High-resolution aerial imagery is now being widely collected and is available for low cost or in some cases is even free. The challenges are to first to align parcel data with the high resolution imagery to identify the exact address and boundaries of a property, and second to develop feature extraction techniques that can exploit the contextual information to accurately identify novel features, such as roofs, cars, pools, landscaping, etc., that can be used for direct marketing. The ability to accurately identify features in imagery and then relate them to specific properties as well as related sources of information will allow a targeted direct marketing product to be built. The end users of this product will be companies seeking to market products directly to residential consumers. This includes product and services relating to home improvement, both exterior and interior, as well as those products relating to residents of the home, that can be gleaned from imagery available for the parcel in question. This is a large market and includes everyone from home improvement stores to roofing companies, construction companies, automobile dealers, tree trimmers, landscapers, and pool construction companies. Beyond direct marketing, the technology can also be used for other applications that combine imagery, geospatial data, and structured information. For example, it could used for mosquito abatement, which is important to stop the spread of West Nile Virus, by identifying large pools of stagnant water, associating those hazards with the appropriate address, and then mailing abatement notifications to the residents.