Ground Truth Validation of Crop Growth Cycle Using High Resolution Proximal and Remote Sensing

地面真相验证（GTV）是成功的特定地点农艺建议（如可变速率处方）的主要组成部分。我们可以通过卫星和无人机图像计算地上生物量（AGB）和产量。然而，这些产品在地面上都不是高分辨率的。目前，农艺师只去有限的几块地，评估不同地区作物和杂草的生长情况，清点种植面积，给作物拍照，并对影响作物产量的因素（如作物种植的一致性）做记录。每年进行一次或两次，以尝试评估当季影响作物的农艺因素。卫星或无人机图像可以显示一个地区生物量低、生长不良的地方，但它不能告诉我们这是由于太干、太湿、盐碱地、贫瘠的植被、昆虫还是生育力差。另一个问题是人工侦察是主观的，耗时和昂贵的。为了解决这些问题，这个研究伙伴关系提出了基于计算机视觉的GTV。为此，近端传感器将安装在农业田间机械上。这些传感器将收集高分辨率图像、土壤电导率、含水量和地形数据。该项目将利用这些近端遥感数据与遥感卫星数据相结合，实现以下四个目标：1)高时空分辨率AGB估计和验证的混合方法；2)均质管理区域的识别；3)作物建立的一致性；4)Kochia杂草管理。

Ground Truth Validation (GTV) is a major component for successful site-specific agronomic recommendations like variable rate prescriptions. We can calculate Above Ground Biomass (AGB) and yield from satellite and drone imagery. However, none of these products are high resolution at the ground level. Currently, agronomists go to a limited number of fields and they assess the growth of crop and weeds in different zones, do plant stand count, take pictures of the crop, and make notes on things affecting crop yield like consistency of crop establishment. This is done once or twice per year to try and assess the agronomic factors influencing the crop during the season. A satellite or drone image can show where an area may be low in biomass and have poor growth, but it does not tell whether it is due to being too dry, too wet, saline, poor plant stand, insects, or poor fertility. Another issue is that manual scouting is subjective, time-consuming and costly. To address these issues, this research partnership proposes computer vision based GTV. For this purpose, proximal sensors will be mounted on agriculture field machinery. These sensors will collect high resolution imagery, soil electrical conductivity, water content and topography data. The project will use this proximal sensing data in combination with remote sensing satellite data to achieve the following four objectives: 1) A hybrid approach for high spatial and temporal resolution AGB estimation and validation, 2) Identification of homogenized management zones, 3) Consistency of crop establishment, 4) Kochia weed management. Advanced machine learning, deep learning and statistical tools will be used to develop novel methodologies. The project will help perform site specific management of crops at the scale of Canadian Prairies. This project will enable the partner organizations collect 20 times more validation samples per field, analyze four times more fields, and cut the costs in half for one million acres of agricultural land. It will also help promote environment friendly agriculture practices in Canada.