植物育種學(xué)家每次會(huì)栽培數(shù)千個(gè)潛力品種；直到現(xiàn)在，對(duì)植物關(guān)鍵特征的觀察都是人工完成的。在一項(xiàng)新的研究中，在對(duì)潛力品種的測(cè)試?yán)?，無(wú)人駕駛飛行器，或無(wú)人駕駛飛機(jī)，可成功地用來(lái)遠(yuǎn)程評(píng)估和預(yù)測(cè)大豆成熟時(shí)間。使用無(wú)人機(jī)來(lái)完成這項(xiàng)工作可以大大減少評(píng)估新作物所需的工時(shí)。

當(dāng)植物育種學(xué)家開(kāi)發(fā)新的作物品種時(shí)，他們會(huì)種植很多植物，而且他們都需要反復(fù)檢查。

“農(nóng)民可能會(huì)有100英畝土地，只種植一個(gè)大豆品種，而植物育種學(xué)家可能會(huì)在10英畝土地上種植1萬(wàn)種潛在品種。農(nóng)民可以快速地確定田地里的單一大豆品種什么時(shí)候才能收割。但是，在秋天，植物育種學(xué)家必須反復(fù)走過(guò)實(shí)驗(yàn)田，以確定每種潛在作物的成熟時(shí)間，” 伊利諾伊大學(xué)大豆育種家布瑞恩 迪爾思解釋說(shuō)。

“我們每三天都必須進(jìn)行檢查，”碩士生內(nèi)森 施米茨補(bǔ)充道?！霸谝荒曛械氖斋@季節(jié)里，這要花費(fèi)我們大量的時(shí)間。而且田地里有時(shí)候很熱，有時(shí)候又很泥濘?！?/p>

為了簡(jiǎn)化工作，一個(gè)跨學(xué)科的研究團(tuán)隊(duì)，包括植物育種學(xué)家，計(jì)算機(jī)科學(xué)家，工程師和地理信息專(zhuān)家都轉(zhuǎn)向無(wú)人駕駛飛行器——俗稱無(wú)人機(jī)領(lǐng)域的研究。

“當(dāng)無(wú)人機(jī)能夠?yàn)槲覀兯?，我們將研究如何才能將這項(xiàng)新技術(shù)應(yīng)用到育種領(lǐng)域。這是首次嘗試，我們?cè)噲D把復(fù)雜的事情簡(jiǎn)單化，”迪爾斯說(shuō)。

其中一個(gè)目標(biāo)是，利用裝載在無(wú)人機(jī)上的攝像頭，以及復(fù)雜的數(shù)據(jù)和成像分析技術(shù)，預(yù)測(cè)蠶豆的成熟時(shí)間?！拔覀兝枚喙庾V成像技術(shù)，”施米茨解釋說(shuō)?！拔覀?cè)诔绦蛑薪⒁粋€(gè)方程式，以便獲取反射在植物上的光頻變化。顏色的變化就是我們?nèi)绾螌⒊墒炫c不成熟植物區(qū)分開(kāi)的依據(jù)。”

研究人員開(kāi)發(fā)了一種算法，將無(wú)人機(jī)獲取的圖像與用傳統(tǒng)方法（通過(guò)田間研究）衡量的蠶豆成熟度數(shù)據(jù)進(jìn)行對(duì)比。我們用無(wú)人機(jī)進(jìn)行的成熟度預(yù)測(cè)非常接近我們田間研究的記錄，迪爾斯指出。

通過(guò)模型做出的預(yù)測(cè)準(zhǔn)確率達(dá)到93%，但是，迪爾斯說(shuō)，如果沒(méi)有無(wú)人機(jī)自身固有的局限性，他們可能會(huì)做的更好。例如，無(wú)人機(jī)只能在陽(yáng)光明媚和風(fēng)力較小的日子里飛行。

對(duì)于它們?cè)谔岣咿r(nóng)業(yè)領(lǐng)域的效率和準(zhǔn)確率方面，無(wú)人機(jī)得到了越來(lái)越多的認(rèn)可，尤其是2016年8月新的FAA（聯(lián)邦航空局）規(guī)則生效后，本研究是首批利用無(wú)人機(jī)優(yōu)化育種實(shí)踐的研究。迪爾斯指出，該應(yīng)用對(duì)于大型育種企業(yè)非常實(shí)用，它們每年要測(cè)試數(shù)十萬(wàn)個(gè)潛在品種。如果利用這項(xiàng)技術(shù)，能夠讓植物育種學(xué)家節(jié)省時(shí)間和精力，新品種就可以被更快地開(kāi)發(fā)出來(lái)供農(nóng)民使用，這是一個(gè)受歡迎的改進(jìn)。

論文，“基于無(wú)人機(jī)平臺(tái)，提升大豆估產(chǎn)方法和植物成熟度預(yù)測(cè)的開(kāi)發(fā)方法”已經(jīng)發(fā)表在《環(huán)境遙感》期刊上。除了迪爾斯和施米茨，Neil Yu, Liujun Li, Lei Tian, 和 Jonathan Greenberg也是該論文的共同作者，他們都來(lái)自伊利諾伊大學(xué)。(張微編譯)

以下為英文原文：

Drones are what's next for plant breeders

Crop breeders grow thousands of potential varieties at a time; until now, observations of key traits were made by hand. In a new study, unmanned aerial vehicles, or drones, were used successfully to remotely evaluate and predict soybean maturity timing in tests of potential varieties. The use of drones for this purpose could substantially reduce the man-hours needed to evaluate new crops.

When plant breeders develop new crop varieties, they grow up a lot of plants and they all need to be checked. Repeatedly.

"Farmers might have a 100-acre field planted with one soybean variety, whereas breeders may have 10,000 potential varieties planted on one 10-acre field. The farmer can fairly quickly determine whether the single variety in a field is ready to be harvested. However, breeders have to walk through research fields several times in the fall to determine the date when each potential variety matures," explains University of Illinois soybean breeder Brian Diers.

"We have to check every three days," masters student Nathan Schmitz adds. "It takes a good amount of time during a busy part of the year. Sometimes it's really hot, sometimes really muddy."

To make things easier, an interdisciplinary team including breeders, computer scientists, engineers, and geographic information specialists turned to unmanned aerial vehicles – commonly known as UAVs or drones.

"When drones became available, we asked ourselves how we could apply this new technology to breeding. For this first attempt, we tried to do a couple simple things," Diers says.

One goal was to predict the timing of pod maturity using images from a camera attached to the drone, along with sophisticated data and image analysis techniques. "We used multi-spectral images," Schmitz explains. "We set up an equation in the program to pick up changes in the light frequency reflected off the plant. That color change is how we differentiate a mature plant from an immature one."

The researchers developed an algorithm to compare images from the drone with pod maturity data measured the old-fashioned way, by walking the fields. "Our maturity predictions with the drone were very close to what we recorded while walking through the fields," Diers notes.

Predictions made by the model achieved 93 percent accuracy, but Diers says they might have done even better without some of the inherent limitations of flying drones. For example, they could only fly it and obtain good images on sunny days with little wind.

Drones are increasingly recognized for their potential to improve efficiency and precision in agriculture—especially after new FAA rules went into effect in August 2016—but this is one of the first studies to use drones to optimize breeding practices. Diers notes that the application could be particularly useful to large breeding companies, which test hundreds of thousands of potential varieties annually. If breeders can save time and effort using this technology, new varieties could potentially be developed and made available to farmers on a faster timeline—a welcome improvement.

The article, "Development of methods to improve soybean yield estimation and predict plant maturity with an unmanned aerial vehicle based platform," is published in Remote Sensing of Environment. In addition to Diers and Schmitz, Neil Yu, Liujun Li, Lei Tian, and Jonathan Greenberg, all from the University of Illinois, are co-authors.

來(lái)源: 中國(guó)科技網(wǎng) 作者: 張微編譯