巴蜀网

 找回密码
 免费注册

QQ登录

只需一步,快速开始

开启左侧
查看: 1162|回复: 0
 彻底的忘记 发表于: 2019-6-13 09:47:03|显示全部楼层|阅读模式

[纪实·新闻] 人工智能助力野保 计数结果快又准

 [复制链接]
  Sue Palminteri | 作者
  朱聪颖、刘润锜 | 翻译
  王聿丹 | 审校
  达成 | 简评
  Monster | 排版
  非洲野保故事
  简评
  近年来,AI人工智能技术占据不少新闻头条篇幅,看似离大众生活很遥远,其实它就在我们身边,如各类美图APPS、汽车自动驾驶技术、网络识图等。如今,人工智能技术也应用到野生动物保护中。
  每年的非洲野生动物大迁徙都是这片大陆上不可多得的壮丽景象,这不仅是以奇景,也是对野保研究的重要课题,迁徙动物种类和数量的多少都会反应该物种的生存情况,过多则会影响草原生态平衡和预示这天敌的减少,迁徙数量过少则反应该物种出于盗猎、食物匮乏等原因。
  传统的计数方法容易出现误差。在塞伦盖蒂和肯尼亚马赛马拉国家保护区之间,有130万头角马和25万头斑马,还没有算上数以万计的角马,动物数量庞大,需要熟练的统计者耗费数周时间进行统计。
  得益于高级算法和民间科学家的支持,一旦学习算法经过学习识别,统计过程所花费时间将大大缩短,统计结果也更加准确。
  文章最后说的“我们研究结果表明深度学习算法可以合法的取代人工计算并为保护组织减轻负担。”看出短时间内学习算法人工智能暂时还无法替代人工计数,但新技术当地潜力是无限的。
  同时也要警惕技术到了不法分子手中,就如公众号里的一篇文章《野保任重道远,网络平台渐为犯罪滋生温床》,技术是无罪的,使用得当,可以拯救无数生灵,若落入居心叵测的人手中,就变成生灵涂炭的屠刀。
  资讯原文
  Migratory species play akey role in the health of the Serengeti ecosystem in East Africa,but monitoringtheir populations is a time-and labor-intensive task。
  迁徙物种对东非塞伦盖蒂(Serengeti)生态系统的健康起着关键作用,但监测它们的种群数量是一项时间和劳动密集型任务。
  Scientists studying thesewildebeest populations compared expert observer counts of aerial imagery tocorresponding counts by both volunteer citizen scientists and deep learningalgorithms。
  研究这些角马种群数量的科学家将专业的观察者对航空图像的计数与民间科学家志愿者和高级学习算法的相应计数进行了比较。
  Both novel methods wereable to produce accurate wildebeest counts from the images with minormodifications,the algorithms doing so FASTer than humans。
  这两种新颖的方法都能够通过微小的修改从图像中得出准确的角马计数,算法计算的的速度比人类快。
  Use of automated objectdetection algorithms requires prior“training”with specific data sets,whichin this case came from the volunteer counts,suggesting that the two methodsare both useful and complementary。
  使用自动目标检测算法需要事先对特定数据集进行“演练”,这些数据集来自志愿者计数,这表明这两种方法既有用又互补。
  A research team testingthe capacity of both citizen scientists and machine learning algorithms to helpsurvey the annual wildebeest migration in Serengeti National Park in Tanzaniafound that both methods could produce accurate animal counts,a boon for parkmanagers。
  一个研究小组测试了民间科学家和机器学习算法的能力,以帮助调查坦桑尼亚塞伦盖蒂国家公园(Serengeti National Park)每年的角马迁徙,发现这两种方法都能计算出准确的动物数量,这对公园管理者来说是一个福音。
  The iconic migration of1.3 million blue wildebeest(Connochaetes taurinus)and 250,000 common zebra(Equus quagga)between Serengeti and the Masaai Mara National Reserve in Kenyais the largest terrestrial animal migration on Earth。
  在塞伦盖蒂和肯尼亚马赛马拉国家保护区(Serengetiand the Masaai Mara National Reserve)之间,130万头角马(Connochaetes Taurinus)和25万头斑马(Equus Quagga)的标志性迁徙是地球上最大的陆生动物迁徙。
c1b6-hyeztys6591070.jpg
  Hundreds of thousands ofwildebeest,plus tens of thousands of common zebra and other grazing antelopemigrate seasonally across Serengeti National Park in Tanzania to find freshgrasses。
  Over one millionwildebeest(Connochaetes taurinus),plus tens of thousands of common zebra(Equus quagga)and other grazing antelope migrate seasonally across SerengetiNational Park in Tanzania to find fresh grasses。
  数以万计的角马,再加上数以万计的普通斑马和其他食草羚羊,季节性地迁徙穿越坦桑尼亚塞伦盖蒂国家公园(Serengeti National Park),寻找新鲜的草源。
  The migration of so manyherbivorous animals affects drives other biological process in the grasslandecosystem,including soil nutrient cycles,the balance of trees and grasses,and the abundance of insects,birds and carnivores。The population trend of thewildebeest in particular reflects levels of bushmeat poaching,disease,andother human disturbance。Understanding the health and dynamics of the migrationis thus of key interest to both researcher and Park managers,yet the sheernumbers of animals have challenged monitoring efforts。
  许多草食动物的迁徙影响着草原生态系统中其他生物过程,包括土壤养分循环、树和草的平衡以及昆虫、鸟类和食肉动物的丰富程度。角马的种群数量趋势尤其反映了灌木丛动物偷猎、疾病和其他人类干扰的水平。因此,对研究人员和公园管理者来说,了解迁徙是否健康和动态是他们最感兴趣的,然而动物的绝对数量却对监测工作构成了挑战。
  “The major driving forcein the Serengeti’s ecosystem is the abundance of wildebeest,”said seniorauthor Grant Hopcraft of the University of Glasgow’s Institute of BiodiversityAnimal Health & Comparative Medicine,in a statement。“[The wildebeest]influence almost every variable in the ecosystem – everything from the returnrate of fires,since they eat the grass,to the amount of insects that areavailable to migrating birds。Without wildebeest,the ecosystem would shiftinto a completely different state,and therefore it’s important to know howmany there are。”
  “塞伦盖蒂生态系统的主要驱动力是大量的角马,”格拉斯哥大学生物多样性动物健康与比较医学研究所的资深作者格兰特·霍普克拉夫特(Grant Hopcraft)在一份声明中说。“[角马]几乎影响生态系统中的每一个变量──从火灾的重现率,因为角马它们吃草,到被迁徙鸟类吃的昆虫数量。没有角马,生态系统将转变为完全不同的状态,因此了解角马有多少是很重要的。”
  Scientists have mostcommonly estimated the population size of these migrating species by flyingaerial transects,taking thousands of photographs,and counting the animalsseen in the images。From these counts,they statistically estimate the densityof animals in the region to come up with an overall population size。
  科学家们最常用的方法是通过飞行空中横断面,拍摄数千张照片,并计算图像中看到的动物数量来估计这些迁徙物种的种群规模。根据这些估计的数据,他们统计地估计了该地区动物的密度,从而得出总体的种群规模。
  However,thelabor-intensive task of manually counting the aerial images can take three orfour skilled counters several weeks to complete,limiting team’s ability tomake a timely,accurate population estimate。
  然而,人工计算空中图像的劳动密集型任务需要三到四个熟练的计数者花费几周才能完成,这限制了团队及时、准确地估计种群数量的能力。
  Producing“highly accurate”counts
  推出“高精确度”记数
  The test showedthat both citizen science and deep learning algorithms can produce accurateimage counts。
  实验表明全民科学和深度学习算法都能推出准确的图像计数。
  As a group,theZooniverse volunteers showed a systematic tendency to undercount the wildebeestin the images,indicating that volunteers were more likely to miss a wildebeestthan incorrectly identify some other animal or object as a wildebeest。
  作为一个团体,动物世界的志愿者总是有少算图片中角马的数量的倾向。比起把其他动物误认成角马,他们更容易漏数。
  In their paper,the researchers suggested that providing volunteers with a field guide toidentifying wildebeest helped prevent overcounting,but that distraction orlosses in concentration that lead to undercounting were more difficult toprevent。The researchers were able to address the undercounts by excluding(filtering)the lowest five of the 15 counts of each image。The average of justthe 10 highest counts closely approximated the expert’s count。
  在他们的论文中,研究者建议为志愿者提供一份鉴别角马的实地指南来预防多数。但是由于注意力分散而导致的少数会更难预防。研究者可以通过排出每张图像的15次计数中最低的5次来解决这个的问题。这样,剩下的10个最高计数的平均值与专家的计算数字就十分接近了。
  They cautioned,however,that this same count bias and the approach that eliminated it mightnot apply to other studies and“there will need to be a rigorousprocess of validation before a citizen science count could be used as the solebasis for a population estimate。”
  然而,他们警告说,同样的计数偏差和消除这种偏差的方法可能不适用于其他研究。而且“在公民科学计数被用作数量预估的唯一依据之前需要通过一个严格的验证过程。”
690d-hyeztys6591109.jpg
  The cumulativeimage counts for three wildebeest counting methods。(a)The mean,median andfiltered mean(just the top 10 counts)for the Zooniverse count data comparedto the expert count。The shaded region indicates the cumulative count thatwould have been recorded if the highest or lowest counts for each image wereused – in other words,substantialerror。(b)The DCNN count compared with the expert count,as its 1.7 wildebeestper image miscount was not systematic in any direction。“A comparison of deeplearning and citizen science techniques for counting wildlife in aerial surveyimages,”published in Methods in Ecology and Evolution。
  累积的图像素材总结了三种统计角马数量的方法:a,平均数,中位数以及筛选过的平均数(只有前十个计数)对于动物世界的记录和专家的记录相比较,打阴影的地区说明了如果每一个图像中将用到最高和最低的计数,也就是说误差较大,那么总的计数将会被记录下来。b,DCNN计数和专家的计数相比较,平均每张图有1.7个角马被数错,这在任何方面都不是系统性的。“深度学习和公民科学技术在航空勘测图像中计数野生动物的比较”,发表于《生态学与进化论方法》。
  With the minormodifications and training,the computer algorithm produced a“highly accurate”wildebeest count,recordinG20,631 animals compared to the expert’s count of20,489。The algorithm did miscount by an average of 1.7 wildebeest per image,though it lacked a systematic counting bias that resulted in a total within onepercent of an expert count。
  经过一些小修改和训练后,计算机算法推出了一个“高精确度“的角马计数法。其记录了20631头角马,而专家的数据是20489头。该算法对每张图像的平均误差为1.7头角马。然而它因为缺少一些系统性的技术偏差,导致了总数不超过专家计数的百分之一。
  Moreover,theresearchers wrote in their paper,it worked far faster than humans can reviewso many images。“The 1,000 images can be processedin under 2 hours,meaning every future census could be counted within 24 hours.Hence,a process that currently takes 3–6 weeks,involving 3–4 wildlife professionals and countless cupsof tea,can potentially be replaced with an automated system that runsovernight。”
  此外,研究人员在他们的论文中写道,它的工作速度远远超过人类对那么多图像的浏览速。这1000张图片可以在两小时内处理完毕。因此,一个通宵运行的自动化系统就可以代替了需要3-6周时间以及无数杯茶来完成任务的的3-4名野生动物专家。
  Complementary methods
  补充的方法
  Nevertheless,DCNNs,like other machine learning algorithms,need to be“trained”to understand the task at hand.Finding sufficiently large training data sets is,the authors write,the“greatest challenge”for implementing thesealgorithms for specific conservation tasks,such as counting wildebeest,penguins,or flowering trees。In this case,the Zooniverse citizen scientistsprovided the necessary wildebeest training data that made the use of thealgorithm possible。
  但是,像其他机器学习算法一样,DCNNs 需要经过“训练”才能理解手头任务。作者说寻找足够大的训练数据集是运用这些算数在特殊的保护任务中的巨大挑战,比如计算角马,企鹅,或开花树木的数量。在这种情况下,动物世界的公民科学家提供了必要的角马训练数据,使算法的使用成为可能。
c7e8-hyeztys6591182.jpg
  Wildebeestmigrating through Serengeti National Park move at various densities。Countingthe individuals in dense groups or reviewing seemingly empty images can be achallenge for non-expert reviewers。
  角马穿过塞伦盖蒂国家公园,以不同密度迁徙。在密集的群组中计算个体数量或者对着看似空白的的图像观察对非专业的评论家来说是个挑战。
  The authors sayin their paper that new data collection technologies,such as camera traps anddrone-borne cameras,might also be able to help scientists build training datasets,expanding the niche of automated image data processing。“Our results show that deep learning algorithms are now at a statewhere they can legitimately replace manual counters and remove a large burdenfrom conservation organisations。”
  作者在他们的论文中写到新的数据收集技术,比如相机陷阱,无人机携带相机,也对科学家们建立训练数据集,扩大自动图像数据处理的市场有帮助。“我们研究结果表明深度学习算法可以合法的取代人工计算并为保护组织减轻负担。”
『 巴蜀网 』提醒,在使用本论坛之前您必须仔细阅读并同意下列条款:
  1. 遵守《全国人大常委会关于维护互联网安全的决定》及中华人民共和国其他各项有关法律法规,并遵守您在会员注册时已同意的《『 巴蜀网 』管理办法》;
  2. 严禁发表危害国家安全、破坏民族团结、破坏国家宗教政策、破坏社会稳定、侮辱、诽谤、教唆、淫秽等内容;
  3. 本帖子由 彻底的忘记 发表,享有版权和著作权(转帖除外),如需转载或引用本帖子中的图片和文字等内容时,必须事前征得 彻底的忘记 的书面同意;
  4. 本帖子由 彻底的忘记 发表,仅代表用户本人所为和观点,与『 巴蜀网 』的立场无关,彻底的忘记 承担一切因您的行为而直接或间接导致的民事或刑事法律责任。
  5. 本帖子由 彻底的忘记 发表,帖子内容(可能)转载自其它媒体,但并不代表『 巴蜀网 』赞同其观点和对其真实性负责。
  6. 本帖子由 彻底的忘记 发表,如违规、或侵犯到任何版权问题,请立即举报,本论坛将及时删除并致歉。
  7. 『 巴蜀网 』管理员和版主有权不事先通知发帖者而删除其所发的帖子。
您需要登录后才可以回帖 登录 | 免费注册

本版积分规则

© 2002-2024, 蜀ICP备12031014号, Powered by 5Panda
GMT+8, 2024-4-18 12:49, Processed in 0.078000 second(s), 8 queries, Gzip On, MemCache On
快速回复 返回顶部 返回列表