SBIR Phase I: Aerial Weed Scout with Robust Adaptive Control for Site-Specific Weed Control

这个小型企业创新研究(SBIR)项目的更广泛的影响/商业潜力是为农民提供按需使用无人驾驶飞行器(UAV)的航空遥感来探测和识别田间杂草的能力。目前，杂草对除草剂的抗药性越来越强。农民被要求花更多的钱，在他们的土地上使用更多的化学物质，导致利润减少和环境退化。与此同时，无人机的成本大幅下降，而美国联邦航空局正在越来越多地放松对飞行的监管。无人机带来了强大的好处：它们可以按需操作(而不是通过载人飞行进行成像需要很长的时间)；与卫星替代方案相比，它们还提供更高的分辨率。然而，农民需要的不仅仅是图像；他们需要可操作的情报，以便做出决定。利用无人机提供的高分辨率图像，以及分析、机器学习和特征匹配方面的进步，这些技术将被用于帮助农民有效地管理他们的除草剂应用，并提高他们的产量。这个SBIR第一阶段项目建议通过开发和测试机载图像处理算法和L1稳健自适应控制来验证空中杂草侦察的可行性，以实现准确、高效的杂草识别和映射。多种技术必须以集成的方式结合在一起，才能将其作为完整的解决方案交付。需要一个稳定的平台，能够在不受风和其他干扰的情况下以极高的精度跟踪飞行路线，以避免因数据收集不准确而可能发生的“垃圾输入/垃圾输出”现象-这对机器视觉算法尤其关键，因为清晰的图像大大提高了精度。关于图像处理和机器视觉，一个关键方面是设计它们，使它们能够在给定机上可用资源的情况下高效执行，并在可用时间内快速执行。目标是以用户友好的方式将这些算法的结果呈现给农民，以便他们可以快速确认或拒绝(视情况而定)结果并采取行动。

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) project is to provide the farmer with the ability, on demand, to use aerial remote sensing via an unmanned aerial vehicle (UAV) to detect and identify weeds in the field. Presently, weeds are increasingly becoming herbicide-resistant. Farmers are required to spend more money and apply greater amounts of chemicals to their field leading to less profit and environmental degradation. In the meantime, UAVs have greatly decreased in cost while the FAA is increasingly loosening regulations to fly. UAVs bring powerful benefits: They can be operated on-demand (as opposed to long lead times for imaging via manned flight); and they also provide much higher resolution as compared with satellite alternatives. However, farmers needs more than images; they need actionable intelligence so decisions can be made. With the high resolution images provided by the UAV, and the advances in analytics, machine learning, and signature matching, these technologies will be used to help farmers efficiently manage their herbicide applications and improve their yield.This SBIR Phase I project proposes to demonstrate the feasibility of an aerial weed scout by developing and testing the onboard image processing algorithm and L1 robust adaptive control for accurate, efficient weed identification and mapping. Multiple technologies must come together in an integrative fashion in order to deliver this as a complete solution. A stable platform capable of following flight paths with extraordinary precision regardless of wind and other disturbances is necessary to avoid the "garbage-in /garbage-out" phenomenon that can occur with inaccurate data collection - this is especially critical for machine vision algorithms as crisp images greatly improve accuracy. Regarding image processing and machine vision, one critical aspect is designing these such that they can be efficiently executed given the resources available on-board, and speedily executed in the time available. The goal is to present the results of these algorithms to the farmer in a user friendly way so that they can quickly confirm or reject (as appropriate) the results and take action.