TUBERSCAN-VENTURE: Delivering a commercially-viable, non-destructive, data driven pipeline to quantify root crops during growth to realise maximum marketable yield and help reduce waste, contributing to net zero emissions

马铃薯产业对英国经济的价值为47亿英镑，但利润率很低，减少农业浪费是一个高度优先事项。马铃薯的大小是至关重要的信息;如果块茎太小，那么它们将不符合客户的规格，而块茎太大则会失去市场价值，这可能是一个巨大的浪费来源。种植马铃薯的农民通常在生长季节对少数植物进行许多破坏性的试验性挖掘，看看他们的马铃薯有多大-然后他们假设他们挖的马铃薯代表了整个田地。然而，大约12%的土豆往往超出了可销售的峰值范围，可能是因为种植面积不同，这使农民每年损失高达1.8亿英镑。在极端天气发生的年份，损失可能会高得多，到2022年，一些田地生产的土豆几乎100%都太小了。我们一直在开发一种技术，可以使用遥感技术收集整个田地的信息，以模拟土豆在地下生长的方式。我们将从相机获得的数据（如植物间距和茎数）与地下对马铃薯质量的感知融合在一起。一个复杂的模型，基于农学知识和参数化的大量背景数据获得了几年来，用于预测整个领域的商品产量。我们已经开发了一个完整的示范模型技术，能够预测商品产量~75%的准确率。我们现在需要优化不同马铃薯品种和土壤类型的数据收集和处理的各种要素。我们还需要确保我们的预测模型在马铃薯生长曲线的所有阶段都是准确的。该项目将使我们更接近这一关键技术的商业化，该技术已应用于全球马铃薯种植以及其他块根作物。

The potato industry is worth £4.7Bn to the UK economy, but margins are tight and reducing farm waste is a high priority. The size profile of potatoes is crucial information; if the tubers are too small then they will not be within specifications for customers, while tubers that are too big lose market value and this can be a massive source of waste. Potato farmers typically undertake many destructive trial digs of a handful of plants during a growing season to see how big their potatoes are getting -- they then assume that the potatoes they have dug are representative of the entire field. Nevertheless, around 12% of potatoes tend to be outside of the peak marketable range, probably because growth varies across a field, and this costs farmers up to £180M per annum. In years where extreme weather occurs, losses can be substantially higher and in 2022 some fields produced close to 100% of potatoes that were too small.We have been developing technology that can use remote sensing to gather information across an entire field to model the way that potatoes are growing while they are still under the ground. We fuse data obtained from cameras, such as plant spacing and stem count, with below-ground sensing of the mass of potatoes. A sophisticated model, based on agronomic knowledge and parameterized with significant amounts of background data acquired over several years, is used to predict marketable yields across the field.We have already developed the technology to a full demonstrator model, capable of predicting marketable yield to ~75% accuracy. We now need to optimize various elements of data collection and processing across different potato varieties and soil types. We also need to ensure that our forecasting model is accurate at all stages of the growth curve of potatoes.This project will bring us closer to commercialization of this key technology, that has application across potato farming worldwide, as well as for other root crops.