New smartphone market: helping farmers cope with crop disease-Smartphone, crop-IT information

A Philippine farmer through his rice field, watching the Orange representing the blast spread on crops. Another peasant of Tanzania in probing of cassava leaves white spots. They look anxious, took out his mobile phone to photograph crop, and soon found what they plant diseases, symptoms and how to treat it.

The above scenario is looking to the future by David Hughes. He is, days closer to the ideal into reality. Hughes and his colleagues Sharada Mohanty, Marcel Salath é, worked in Switzerland the Federal Polytechnic of Lausanne. They imported in a computer image from the 54,000 plants leaves, and let the computer society based on leaf characteristics to determine what to plant diseases. In this process using computer algorithms, can be diagnosed 14 of 26 different crops diseases, accuracy rate of up to 99.35%.

It should be recognized that the basis of all this, is in bright light conditions and eligible under the background of photos of the plants. Because of the need to meet the conditions of this idealism, this algorithm does not say perfect. However, the development team thought it was irrelevant, but needs a large number of selected pictures .

"It has proved, our computers may automatically according to the status of the surface of the leaves differentiates very complex pathology," Hughes said, "no one else in the world can do that. If you are a tomato experts in the field, you treat your tomatoes at best, but we can to treat 26 plants, you can't. ”

Hughes knew plant diseases. He was in Ireland Dublin, grew up, eyes see is a barren mound. But even the land of potatoes are grown Fusarium, 150 years ago that the disease is caused by famine and social upheaval (translator's Note: the 1845-1852 Ireland potato late blight oomycete and crop failures and famine caused the country and more than 1 million Ireland people died). As an adult, Hughes worries gradually, plant disease not concerned on the one hand, on the other hand can lead to disaster. 90% world of calories provided by 15 crops, these crops are mated with a single species, are prone to diseases.

? Infection by late blight, potato profile

Due to various reasons, plant pathologist at plant infectious diseases rapid reduction in the number of. When experts retire, their knowledge is being sealed up in the mind.

In order to reduce the loss caused by insufficient experts, Hughes and Salath é launched PlantVillage website. You can think of it as a botanical version of Quora: people around the world are available on the PlantVillage to ask questions, seek advice from farming. Thus, agricultural knowledge farmers can understand each other, to make up for the insufficient number of experts. This is a good idea, but despite the increase in Web site traffic steadily, most questions are answered by the PlantVillage team.

So, Hughes and others began to look for another way out, to use a computer to automatically diagnose plant diseases. First of all, they need to plant photos. So they went to Ghana's experimental farm: where researchers to vaccinate their crops of selected pathogens for testing. By this way, researchers will be able to fully determine their photographs in the plant, is what ill. Hughes, who shot totals about 100,000 photographs, all taken in the sunny day, photo background is uniform .

Then, Salath é was responsible for half the amount of photos to import an artificial neural network. This is a special computer system to simulate a form of communication between neurons in the brain, and learning how to identify and classify a bunch of pictures. The network uses some pictures of crop leaves, has developed a program that can identify these leaves are what the illness, then the other leaf photo test the program. As mentioned earlier, which made 99.35% accuracy, satisfactory.

In addition to Hughes's team, and other research teams have tried a similar approach, but according to the Salath é said, those teams can only solve a particular crop or a disease: "our journey is 14 and 26 species of crop disease. We will not inform the computer photographs of the plants with Phytophthora infestans, also it is not prompted to check the grey spots. We just allow the system to network their probes, to figure out which pattern implies plant diseases. ”

Run the neural network requires massive computing power, but once the program was established, is sufficient in the Smartphone. In addition, since the world's population on the rise--the proportion of mobile broadband access reached 69% by the end of last year, the team optimistic expectations and growing--Hughes, most farmers in developing countries need to diagnose plant diseases can also detect plant disease using mobile applications.

Of course other mobile application has been developed to diagnose plant diseases, but they were either given a lot pictures of plants to the farmers themselves, or build a bridge between farmers and agricultural experts. Mobile application of automatic diagnosis of crop diseases, only this one.

However, the United States Purdue University (Purdue University) Janna Beckerman doubts this auto diagnose crop diseases how useful mobile applications can. "Diagnosis of plant diseases, I cannot only look at the leaves and appearance," Beckerman said, "I need a microscope and meticulous checks. Sometimes I still need to try planting it. This is the mobile application, but we have to remind people, should follow the laboratory diagnostic results.

United States Agency for international development, Judy Payne's response more enthusiastic: "help farmers diagnosis tricky crop diseases, potential impressed me in this area. Although technology is not enough to replace the experienced plant pathologists, but it can drastically reduce the challenge faced by farmers in developing countries, both at a distance and is a resource. ”

In addition, Payne added: " this technology needs a good test, to adapt to the environment of the rural areas. "Indeed, the Hughes team more realistic pictures of, such as books, scientific articles and other sources to obtain photos, diagnostic accuracy rate fell only 31%. When the program is informed that the crop species, accuracy rose to 45%.

This data isn't impressive sounding but Salath e pointed out that, if there is any random guessing, the program correctly only 2%. "The program is better and more", Hughes said, "who can distinguish between 14 26 kinds of crop diseases, accuracy is higher than 45%? I think not. If a Department includes experts from different fields of expertise, the sector as a whole may be able to achieve the accuracy of 45%, but our method allows United Kingdom of Kent County, United States, Kentucky, and Kenya's farmers can achieve diagnostic accuracy of 45%. ”

Hughes's team open their photo database, and look forward to computer experts to help them design better algorithms. Their efforts to collect more photos of crops. These crops were erosion and infection of the disease cycle and photographed in a bright background light. Salath é pointed out that this is the whole point: "at present, our algorithm is not sufficient to distinguish the difference between so many pictures. We now need to do, is to add more photos, this program automatically diagnostic accuracy rate will go up. ”

Hughes has several photography team, respectively, in photos taken in Tanzania and the photo shoot of rice in the Philippines, and cassava. James Legg is part of an international agricultural research organization (the Consultative Group for International Agricultural Research,CGIAR), is also a partner in Hughes. Legg said: "I am mainly working in Tanzania. This technology allows tens of millions of farmers so their crops from Mosaic and brown streak of attacks. You know, these diseases will cost Africa more than US $ 1 billion a year. ”

"Our goal is to take 3 million pictures in three years," Hughes said, "our photos will be free and open to the public. If you want to use our pictures, your account must be open access to others. ”

"I can see the world's farmers use crop considerable prospect of application of automatic diagnosis of diseases," the John Innes Centre (John Innes Centre) Saskia Hogenhout said: "in addition, I am sure that the incidence of people volunteered to help take crops and pests photos to enrich public database. This will help the plant quarantine agencies around the world to control crop diseases and pest attacks. ”

智能手机新销路：帮农民应对作物灾病 - 智能手机,农作物 - IT资讯

一位菲律宾农夫穿行过他的稻田，看着代表稻瘟病的橙色蔓延上庄稼。而另一位坦桑尼亚农民则在探查木薯叶子上的白色斑点。神色焦急的他们，拿出手机给作物拍照，并很快就找出了他们的植物生了什么病、症状如何，以及该怎么治疗。

上述场景正是David Hughes所展望的未来。令他兴奋的是，理想照进现实的日子更近了。Hughes和他的同事Sharada Mohanty、Marcel Salathé，都任职于瑞士洛桑联邦理工学院。他们在一台电脑中导入了5.4万片植物叶子的图像，并让电脑学会依据叶子的特征来判断该植物生了什么病。在这一过程中使用的计算机算法，可以诊断出14种作物的26种不同疾病，准确率高达99.35%。

应当承认，实现这一切的基础，是要在明亮的光线条件及合乎标准的背景下拍摄出植物的照片。由于必须满足这种理想主义的条件，因此这种算法还不能说完美无缺。不过，研发团队认为这倒无关紧要，只不过是需要大量选取图片罢了。

“现在已经证明，我们的电脑可以根据叶子表面的状态自动区分非常复杂的病理，”Hughes说道：“世界上还没有其他人能做到这一点。如果你是西红柿领域的专家，你顶多只能给西红柿治病，但我们能给26种植物治病呢，你就不行了吧。”

Hughes很了解植物的疾病。他在爱尔兰的都柏林长大，满眼看到的都是贫瘠的土堆。这可是连土豆都能种出枯萎病的土地啊，150年前的那场灾病更是导致了大饥荒和社会剧变（译者注：1845~1852年爱尔兰土豆因晚疫病的卵菌而歉收并造成全国范围内的饥荒，100余万爱尔兰人因此死去）。作为一名成年人，Hughes逐渐忧虑，植物疾病一方面不受关注，另一方面又会带来灾难。全球大约90%的卡路里是由15种作物提供的，而这些作物都是采取同系交配的单一品种，很容易产生疾病。

▲感染了晚疫病的土豆剖面

由于各种原因，擅长解决植物传染病的植物病理学家的人数在迅速减少。老专家退休后，他们头脑中的知识也就被尘封起来了。

为了降低专家不足带来的损失，Hughes和Salathé建立了PlantVillage网站。你可以把它视为植物学版本的Quora：全世界的人们都可以在PlantVillage上发问，寻求农业种植的建议。由此，农民们可以互相了解农业知识，以弥补专家人数不足的情况。这无疑是一个好主意，但是尽管网站流量在稳步上涨，大多数问题还是由PlantVillage团队自己回答的。

所以，Hughes等人就开始另谋出路，以期用电脑自动诊断植物疾病。首先，他们需要植物的照片。因此，他们前往位于加纳的试验农场：在那儿，研究人员给作物们接种一些选定的病原体以进行试验。通过这样的方式，研究者就能完全确定，他们拍摄的照片中的植物，是生了什么病。Hughes等人总计拍摄了约10万张照片，全都是在阳光充足的白天拍摄，照片背景也是统一的。

接着，Salathé负责将一半数量的照片导入一个人工的神经元网络。这是一种特殊的电脑系统，能模拟出大脑中的神经元之间的交流形式，并习得如何给这么一堆照片进行辨别和分类。该网络使用了一些作物叶子的照片，编制了一个能够辨别这些叶子都得了什么病症的程序，再用其他的叶子照片测验这个程序。如前所述，该程序取得了99.35%的准确率，令人满意。

除了Hughes的团队外，还有其他研究团队也尝试了类似的方式，但是据Salathé说，那些团队只能解决某一种作物或是某一种疾病：“我们的征途是14种作物和26种疾病。我们不会事先告诉电脑，照片中的植物是否带有疫病菌，也不会提示它检查那些灰色斑点。我们只是让系统网络自己探查，找出哪些特征预示着植物的疾病。”

运行上述的神经元网络需要占用海量的运算能力，但是一旦程序建立起来了，在智能手机上运行就足够了。此外，自从全球人口的移动带宽接入比例在不断上升——去年年底达到了69%，而且还在不断增长——Hughes的团队乐观期望，最需要诊断植物疾病的发展中国家农民，也可以用手机应用探查植物疾病。

当然其他诊断植物疾病的手机应用已经被开发出来了，但它们要么是给出大量的植物图片给农民自己比照，要么是搭建农民和农业专家联系的桥梁。自动诊断作物疾病的手机应用，还只有这独一份。

不过，美国普杜大学（Purdue University）的Janna Beckerman怀疑，这款自动诊断作物疾病的手机应用到底能有多大用处。“要诊断植物疾病，我不能只看叶子和外表特征，”Beckerman说道：“我需要用显微镜细致检查。有时候我还需要试种植一下。这只是手机应用而已，但我们一直都提醒人们，应该听从实验室的诊断结果。

而美国国际开发总署的Judy Payne的反应要更热情一些：“帮助农民诊断繁难的作物疾病，这个领域的潜力让我印象深刻。尽管现在的技术还不足以替代经验丰富的植物病理学家，但能够大幅减少发展中国家农民面临的挑战，不论是在距离上还是资源上。”

此外，Payne还补充道：“这门技术需要好好测试，以适应乡村地区的环境。”确实，在Hughes的团队用更符合现实的照片的时候，例如从书本、科学论文和其他来源处获取照片时，诊断的准确率就跌倒仅有31%。而当程序被告知作物种类时，准确率能回升到45%。

这个数据听上去虽然不怎么样，但Salathé指出，如果是随机猜测的话，程序的正确率只有2%。“程序最好还是和人比较”，Hughes说道：“难道有谁可以在区分14种作物的26种疾病方面，准确率要高于45%吗？我看不行。如果由部门包括了不同专业领域的专家，那么这个部门作为整体而言或许能达到45%的准确率，但是我们的方法，可以让英国肯特郡、美国肯塔基州和肯尼亚的农民都能达到45%的诊断准确率。”

Hughes的团队免费开放了他们的照片数据库，并期望计算机专家能帮助他们设计出更优秀的算法。他们努力收集更多的作物照片。这些作物被各种疫病侵蚀且处于不同感染周期，并且在光照明亮的背景下拍下照片。Salathé指出，这正是关键所在：“当前，我们的算法还不足以区分那么多照片之间的差异性。我们现在需要做的，就是添加更多的照片，这样程序自动诊断的准确率就会上去了。”

Hughes有几个摄影团队，分别在菲律宾拍摄水稻照片以及在坦桑尼亚拍摄木薯照片。James Legg隶属于国际农业研究磋商组织（the Consultative Group for International Agricultural Research，CGIAR），同时也是Hughes的合作者。Legg表示：“我主要在坦桑尼亚工作。这项技术让好几千万农民受益，使他们的作物免受花叶病和褐条病的侵袭。要知道，这些疫病每年都会让非洲损失超过10亿美元。”

“我们的目标是在三年内拍摄300万张照片，”Hughes表示：“我们会将照片免费开放给公众。如果你想使用我们的照片，你的账号也必须是向他人开放获取的。”

“我能看到全球农民都使用作物疾病自动诊断应用的可观前景，”约翰英纳斯中心（John Innes Centre）的Saskia Hogenhout说道：“此外，我确信，人们会志愿帮助我们拍摄发病作物以及害虫的照片以充实公开数据库。这将帮助世界各地的植物检疫机构控制作物疫病和害虫侵袭。”