1. The lightness of life
Chu Wenwen knew from a young age that one day, no matter how far she went, she would return to Xinjiang.
Later, she went to university. After four years of undergraduate studies in Beijing, many of her classmates chose to stay and become a North drifter, but Chu Wenwen packed her bags and returned to Xinjiang without any hesitation.
Caption: Chu Wenwen in Altay
The distance from Beijing to Altay, Xinjiang is about 3,952.49 kilometers, which is similar to the distance from Shandong to Xinjiang (3,841 kilometers) by her father in the 1990s, which is nearly 4,000 kilometers.
She remembered that her father, Chu Hongjun, came to Xinjiang as soon as he graduated with a master’s degree. Because he studied beavers during his master’s degree, he was assigned to the wild beaver protection station in the Altay area to engage in scientific research work. The first group of grass-roots wildlife protection researchers after the Wildlife Protection Law.
Altay is located in the northern part of Xinjiang, at the southern foot of the Altai Mountains and adjacent to the Junggar Basin in the south. In 1990, the number of beavers in China was only about 500, which is rarer than the giant panda. In nature, when the number of a species is less than 400, it will irreversibly move towards inbreeding, the race will disappear, degenerate and even die out. If humans do not intervene, the disappearance of beavers on Chinese soil will be an irreversible end. Therefore, the arrival of the early Red Army has a Taishan-like significance to Beaver.
When she was two years old, Chu Wenwen also came to Altay with her mother to support her father’s wildlife protection work.
Because of the shortage of manpower, Chu Wenwen’s father took her to work with her when she was very young: when she was seven years old, the Chu Red Army taught her to ride horses, use the transect method to survey wild animals, and make tabular data in the computer input of. When he went out for inspections, he would also take Chu Wenwen by his side, no matter the mountains and grasslands, no matter the cold wind or heat, Chu Wenwen also began to use binoculars to look at distant animals from a young age, and use the camera to see wild animals. Filmed, recorded and archived.
What impressed Chu Wenwen even more was that her father went home to rescue injured wild animals. When she was very young, the Chu Red Army once brought a little yellow sheep to her home that had a car accident. The lamb was bleeding and whimpering, and her life was dying. This brought a great shock to Chu Wenwen’s young mind:
“Because I grew up in Altay, there are many wild animals in Altay, and I will acquiesce in my heart that they all grew up with me.”
But the cruel side of nature also unleashes its lethality. In addition to car accidents, a wild animal may encounter many unpredictable and unavoidable injuries in harsh wild environments. For example, a wild fox may die by eating poisoned mice just in the process of foraging, not to mention habitat degradation and other more irresistible external factors.
To save, or not to save? How to save? This is a question that Chu Wenwen has been thinking about since she was a child.
Almost every winter, her father would take her to feed the frail goose antelope, wild horses, and wild asses that were too thick to eat because of the snow. Altay’s winter can be as low as minus 40 degrees in the coldest time, and a trip often takes seven or eight hours in the wild.
Caption: Winter in Altay
When casualties occur in a large population, the disappearance of life is often attributed to the advancement of natural laws, but for some endangered species, the death of a life is a matter of how far a population can go in the long river of human history. For example, there are only about 500 beavers in China. If their overall reproduction number is lower than the death number for a period of time, it means that the Chinese beaver is getting closer and closer to extinction.
For beavers, being “injured” is a luxury.
Therefore, Chu Wenwen was determined to be a wildlife scientist like her father since she was a child. As an undergraduate, she took a double major in English and media at China Agricultural University to prepare for reading a large number of foreign literatures required for research on wildlife. After graduation, she was admitted to Beijing Forestry University for postgraduate and Ph.D.
In the summer of 2017, after graduating from her undergraduate degree, Chu Wenwen returned to Xinjiang. The proportion of classroom teaching in the wildlife major is very small, and more is field research and practice. Altay has more than 400 species of wild animals, 74 of which are listed as national protected animals, which is a unique experiment for her. room. In the second year, she initiated the establishment of the Altay Regional Nature Conservation Association (hereinafter referred to as the “Association”), dedicated to the practical protection of wild animals, among which the beaver is the key target protection target of her inheritance from her father.
2. The value of records
How to protect beavers? This is a huge problem.
In the process of doing research with her father since she was a child, Chu Wenwen knew that there are 197 broods of Mongolian and New Beavers in my country, mainly distributed in the Ulungu River Basin in Altay. The Ulungu River is 725 kilometers long, which is about one-fifth of the distance from Beijing to Xinjiang. Whether it is support or rescue, it is not easy to carry out work.
Caption: Ulungu River Basin
Chu Wenwen and her team have done a lot of work in the early stage. Among them, the most extensive work is to set up more than 500 infrared cameras along the 725-kilometer-long Ulungu River, mainly to obtain real-time information that is beneficial to beaver protection. Survival data, including:
The impact of different seasons, upstream and downstream river courses, river droughts and floods, and human activities on the survival of beavers;
The impact of beavers on the surrounding ecological environment, such as whether more species gather in the niches made by beavers, and whether the number of species gathered in the niches made by beavers increases every year;
Beaver reproduction data, such as fertility status and cub survival rate, when the camera can fully cover each beaver litter, by photographing and analyzing the surrounding ecological environment factors, more places that can help beavers to increase their population can be found. .
These image data play a very important role. Association members Wang Dapeng and Jin Bochen both sighed: “Although the acquisition and analysis of image data is a preliminary work, it is at least 50% important. Without the support of these data, we would not know where to go if we had the strength.”
In other words, to fight the battle to protect the beaver, the personnel have arrived, and the strength and faith have also been mustered, so what is the goal? The guidance of the image data captured by these 500 infrared cameras is needed. This cannot be done by imagination alone, nor by experience alone. For example, in 2019, Chu Wenwen and her team discovered in the early analysis of scientific research data and image data that the most urgent thing for beavers to solve is how to have sufficient new food resources and provide habitat and shelter for the beaver population. conditions for living habitats. On the basis of scientific research data, the association decided to launch a public welfare project called “Beaver Canteen” to plant shrubs and willows in more natural habitat factors close to the beaver’s habitat, and artificially restore the habitat environment for the beaver, thereby Solve the problem of beaver’s feeding and breeding environment.
But here’s the thing: there are 500 IR cameras in total! It is conceivable that the amount of video data recorded and photographed in one year is huge, and how to efficiently process this data has become the primary problem in the protection of beavers.
“The data sources include surveillance video/infrared camera collection and field observation and shooting. We have calculated that the amount of data we generate in a year exceeds 400 Tb. If we want to fully digest these data, it will take thousands of hours of manual work. Usually, the data from last year has not been digested, and a lot of data is accumulated in the new year. Faced with this situation, we are very helpless.” Jin Bochen introduced.
Moreover, they found that relying on the human eye to identify the activity frequency of beavers, the number and types of other creatures around the beaver den, etc., by the human eye is tantamount to finding a needle in a haystack, which is not only slow, time-consuming, and difficult, but also Very low accuracy.
Caption: The first photo of a beaver in China building a dam in the wild, taken by Chu Wenwen
Is there any way to solve this problem?
In 2020, by chance, Chu Wenwen and members of the association saw in the news that Chinese scientists used AI technology to identify different giant panda individuals, and the effect was very good. Problems with image data processing. With this idea, they started looking for a team that could provide “X-face” recognition technology.
At this time, the Beijing Baller-Tech team, which is 4,000 kilometers away, is also looking for a scene to implement AI algorithms in wildlife protection.
The Baller-Tech team consists of 5 young people: 1 data engineer, 2 algorithm engineers, 1 software development engineer and 1 product manager. All of them have just resigned from the company and are preparing to start their own businesses.
Among them, algorithm engineer Cao Ji was originally the chief technology officer (CTO) of an artificial intelligence company. He has worked in the field of visual algorithms for more than 20 years and has profound experience in landing research. At the same time, he is also a wildlife enthusiast, Environmentalists have always hoped to use artificial intelligence to help improve the natural ecological environment.
Legend: Cao Ji
While the Altay Regional Nature Conservation Association was “looking for someone”, Baller-Tech was also looking for a scene.
Finally, Chu Wenwen found Baller-Tech through a friend who is also enthusiastic about nature and environmental protection. At that time, Baller-Tech was already interested in the field of ecological protection and had already carried out tentative research, so they hit it off just through a remote Tencent meeting.
Baller-Tech understands AI, and Chu Wenwen has a deep understanding of wildlife protection. The two sides complement each other’s advantages, which is obviously a win-win cooperation.
3. Difficulties in the intersection of parallel lines
“It’s definitely a beaver!”
“This is definitely not a beaver!”
Jin Bochen recalled and smiled helplessly: During the entire product development process, these were the two sentences they heard the most.
Generally speaking, giant pandas are large in size and simple in activities. The image collection of giant pandas can often get clear and accurate pictures, but the gestures and behaviors of beavers living in wild waters appearing in the camera are much more complicated. For example, sometimes the beaver only exposes a small part of its body in the water, and the features of this part of the body are not so obvious that it is difficult for even a professional to judge, let alone a technician who does not know much about beaver.
“We’re always blushing. The point is that in the area where the beaver lives, there is often a muskrat that looks like it. This thing should make our data engineer’s hairline move up another centimeter. …” the Baller-Tech team laughed.
Beavers are known as “ecologically oriented species”. They build their own dams to form a small pond and feed some water plants and algae with their feces. With aquatic plants, fish will be attracted to gather, and fish will attract water birds, foxes, golden eagles, badgers, etc., wherever there is a beaver nest, it will become vibrant.
In addition, the living environment of the beaver is relatively complex. There will be grass, shrubs and trees in its range, and it will be blocked when it moves. At the same time, the beaver is a nocturnal animal, and the collected images will be dim and blurry. In many cases, the color of the beaver is also mixed with the surrounding environment, making it difficult to distinguish. These all pose considerable challenges to observation and data labeling in the early stage, and are not conducive to the optimization of AI algorithms.
Legend: Beaver photos collected by infrared cameras
Artificial intelligence and wildlife protection seem to be two parallel lines that do not intersect at all, but on the goal of beaver protection, the Altay Regional Nature Conservation Association and Baller-Tech have a wonderful encounter, and they have reached an agreement: it is to Introducing AI to the beaver conservation gigantic project to help solve some of the problems.
However, in the early communication, the two sides inevitably encountered a typical interdisciplinary cooperation problem: what does one party really want, and what can the other party give?
“Our environmental workers need to understand that AI technology doesn’t give it a photo, and it can accurately find the beaver in the photo. And technology workers also need to understand why the beaver can’t do it quickly like the giant panda. Accurate individual identification,” the association members explained to us.
In order to achieve the unification of scientific research paths, the Baller-Tech team and the association hold meetings twice a month, and also travel from Beijing to Xinjiang three times to conduct on-site inspections to understand real application scenarios and conduct environmental analysis.
Caption: Baller-Tech representatives met with Chu Wenwen in Xinjiang
Finally, they settled on the route, which is divided into three steps:
One is to develop a special labeling tool for the massive image data provided by the association; the other is to separate the content captured in the massive images from the images that did not capture the content, because up to 90% of the data collected by the infrared camera are No practical content has been collected; the third step is to accurately label the collected content, and then train the model to identify beavers and other species.
In the process of cooperation, Baller-Tech found that although many AI algorithms currently have open source models for reference, and most open source tools are universal and can be used to verify ideas and solve some problems, but for In the serious and professional field of ecological protection, from “running” to “available” is often separated by layers of barriers, requiring a lot of adjustment and optimization, or even overhauling it. After developing specialized annotation tools, their efficiency has improved a lot.
But in addition to data labeling and model optimization, their work also faces an urgent problem: computing power.
On the one hand, computing power is a high-cost investment for Baller-Tech and the association. Environmental protection is not an 800-meter sprint, but a marathon walk. In addition to the processing of image data, there are also many branch jobs. For example, Chu Wenwen and members of the association launched “Beaver Canteen”, “Beaver Guardian”, “Beaver” and “Beaver” on Tencent Public Welfare. The three projects of Beaver Ark are planting badger willows for beavers to eat, giving financial allowances to local herdsmen participating in beaver inspections in Altay, and building a rescue center for injured beavers and other wild animals. For the project, these are not small expenses.
On the other hand, solving computing power requires deploying servers. The Baller-Tech team is mainly stationed in Beijing, and the Altay Regional Nature Conservation Association is located in Xinjiang. It is a question of which city the server is located in. If it is placed in Xinjiang, the energy of Baller-Tech’s remote deployment and maintenance will be increased, and the network in the wild in Xinjiang is not stable enough, and the process of transmitting data from Xinjiang to Beijing through the network will be very slow.
Due to the limitation of computing power, although they have achieved many good results in the past three years, the progress is still relatively slow.
4. AI on the cloud
While they were looking for ways to solve the technical bottleneck, the three projects initiated by the association and Tencent Philanthropy also achieved good results one after another. This gave them a lot of encouragement, because public welfare is, by its very nature, something that is sustained by pure love.
From 2018 to the present, in more than four years, the “Beaver Canteen” project has successfully planted more than 600,000 badger willows in the Ulungu River Basin, solving the food problem of beavers. It’s not an easy task:
“Because the bank of the Ulungu River is sandy soil, which is equivalent to planting trees on the sand, the water content itself is very low, and the sun on the Gobi Desert in Xinjiang in summer is very strong. The river is 750 kilometers long, and we are hiking. I walked with a bucket to water the tree. When I planted the tree, it was often watered in the morning, then dried up at noon, and wilted in the afternoon. Everyone worked hard and got tanned, but the survival rate of the saplings was still very low.” Chu Chu Wenwen recalled the scene of the first year.
Caption: Chu Wenwen and members of the association inspected the growth of badger willow near the Ulungu River Basin
Later, Chu Wenwen returned to her alma mater, found professors from Beijing Forestry University, and asked them about the plan for planting trees. She tried several planting methods in the second year, and only broke through the difficulty of planting trees in the second year. The survival rate increased to 50%, 60%, and then to 70%, 80% in the third year.
Food alone is not enough, they also called 500 herders to act as rangers along the 750-kilometer Ulungu River Basin. The local Kazakh herdsmen speak Kazakh language, and Chu Wenwen and the team members started to learn Kazakh language from scratch, not only speaking but also writing. , basic course training in botany, entomology, ecology.
Through their efforts, in four years, the number of endangered beavers has increased from about 500 to 610, an increase of 20%. It was called a “miracle” by the United Nations and was selected as one of the 100 classic cases of biodiversity conservation in the world at the United Nations COP15. This year, Chu Wenwen also won the highest honor for Chinese youth – the “May Fourth Youth Medal”.
These results are a great encouragement to both the Altay Regional Nature Conservancy and the Baller-Tech team, both affirming their efforts and driving them to continue to push forward, race against time, and within a limited time Do more to protect wildlife.
At the beginning of this year, members of the association noticed that Tencent cooperated with China Children’s Center and was initiated by Tencent Penguin Partners Growth, Tencent Sustainable Social Value Division Innovation School Lab, Tencent Penguin Love the Earth, Tencent Youtu Lab, Tencent Cloud AI, Tencent Cloud The 2nd “Tencent Light Public Welfare Innovation Challenge” (hereinafter referred to as “Light Challenge”) jointly organized by Weidai Low Code and Tengyuan Association.
The Baller-Tech team was originally worried about computing power, and always wanted to find a stronger AI team to support. They learned that Tencent Cloud, relying on top AI research teams such as Tencent Youtu Lab, has achieved very good results in the development of AI technology on the cloud. , we clapped our hands at the moment and decided:
“We must participate. We are not for anything else, but if technical research and development can get technical support from Tencent Cloud, our project can move forward quickly.”
It is understood that Tencent Cloud AI processes over 5 billion images per day, over 2.5 million hours of speech per day, over 100 billion sentences per day in natural language, and serves over 1.2 billion users worldwide. In this year’s Tencent Light Challenge, Tencent Cloud AI, together with Tencent Youtu Lab and Tencent Cloud Microbuild, provided participants with a variety of AI technologies and a low-code platform for microbuilds, attracting more than 1,438 participating teams.
Since going to the cloud means the disclosure of data, no non-profit organization in China has tried to put data on the cloud before.
From the perspective of technical effects, cloud services have natural advantages, which are high efficiency and convenience. Jin Bochen told AI Technology Review that their participation in the Tencent Light Challenge this time was also a bold attempt, but the benefits of participating are also obvious: one is to solve the bottleneck of computing power, and the other is that after going to the cloud, Baller-Tech can By deploying models on Tencent Cloud and providing services through Tencent Cloud, the Altay Regional Nature Conservation Association can also directly call and analyze data from Tencent Cloud.
Tencent Cloud provides a one-stop service for developers to deploy, expand, monitor, and display, so that Baller-Tech team members can focus more on Beaver’s AI core research and development.
After going to the cloud, they used to manually identify things that took two to three months, or even half a year, but can now be completed in two days.
The power of technology shocked Wang Dapeng.
He joined the association last summer and just witnessed the association’s precise process of beaver protection from manual identification to machine identification and invoking Tencent’s AI technology, which refreshed his understanding: “If there was no support for Tencent’s AI technology, we would The efficiency of nature protection work in the Altai Mountains is still quite poor, but it is different now, and our results can be reported to the United Nations.”
Therefore, “Tencent Cloud’s AI support is not very important, it is very important, and it is one of the most important!”
With the help of Tencent Cloud, they can now quickly receive data feedback and grasp key information, such as understanding whether the annual restoration of beaver habitat is in place, how effective it is, whether new species have increased, and the local biodiversity Sex has not changed, animals have not lost, and so on. This information can be quickly analyzed.
According to him, after they achieved such excellent results using AI technology on the cloud in beaver protection, their counterparts across the country are now looking to them for advice on how to introduce AI into ecological protection. This is something they didn’t expect: “This shows that this technology can be extended to deep mountains and old forests all over the country to identify other species.”
5. Broader exploration
Although the image recognition technology used in beaver protection is a very mature research in the field of AI, it still has excellent application scenarios that affect a larger society and more groups living on the earth, in the second Tencent Light Public Welfare. Stand out in the Innovation Challenge and won the first place in the competition. Leifeng.com
Wu Yunsheng, vice president of Tencent Cloud and general manager of Tencent Youtu Lab, told AI Technology Review that since Tencent started the “930” reform on September 30, 2018, “self-developed cloud” has become one of the two major directions for the development of Tencent’s technology system. (The other is “Open Source Collaboration”), as AI researchers, they are responsible for the research and implementation of AI algorithms; and in the Tencent Light Public Welfare Innovation Challenge, they hope to see more contestants and developers can learn from social life. Identify problems and solve them through technical capabilities such as AI provided by Tencent Cloud.
In April 2021, Tencent established the Sustainable Social Value Division (SSV), which is committed to technological innovation to jump out of the small circle of helping the development of the Internet economy and move towards greater social sustainable development, including basic science, educational innovation, rural revitalization, and public welfare. funding, etc. In the Light Challenge, the combination of scenes such as beaver protection and AI verifies this.
The addition of AI on the cloud has brought more imagination to Chu Wenwen and others to promote wildlife protection work, but just as virtual currency is to blockchain, image recognition is to the content that AI can participate in wildlife protection. It’s just the tip of the iceberg, or “a drop in the bucket”.
Chu Wenwen told us that, 365 days a year, they spend at least 300 days in the wild. In order to photograph a beaver, she lay motionless for eight hours a day in the snow at minus ten degrees for a month.
They have done a lot of things, planting trees for beavers, raising donations for beavers, having more than 400,000 followers on Douyin, getting more than 2 million likes, and building rescue centers for beavers and other wild animals – “Beavers” The Ark” project – is also about to land. In the past four years, they have rescued 76 injured wild animals, in addition to beavers, there are snow leopards, lynx, golden eagles, vultures, goose antelope and so on.
In this process, in addition to physical torment, there is also emotional torment: “We did our best to save many wild animals, but we couldn’t save them. When your team put a lot of effort into doing this, In the end, it left, which is a very serious emotional blow to people.” Leifeng.com
In four years, the beaver has indeed grown by more than a hundred, but it is still a long way from breaking the endangered line. Wang Dapeng candidly said that it took nearly 30 years for the predecessors of wildlife protection (such as Chu Wenwen’s father, Chu Hongjun) to keep the beaver from disappearing. Knowing that there are many obstacles in the way, “perhaps by the end of our careers, we may not necessarily see the beaver downgraded from the first-level national protected animals to the second-level and third-level protected animals.”
But the emergence of AI still gave them great expectations.
The Baller-Tech team is now also a member of the association. In the process of this beaver cooperation, they also discovered a wider application of AI technology in wildlife protection. For example, in addition to image recognition, in order to facilitate the team to carry out scientific research and protection work in the Altai region, they also “smoothly” developed AI language recognition and translation tools for Kazakh and Uyghur.
This was not in their plan, but unexpectedly turned into a willow tree, opening up a rich space for them to explore the combination of AI and ecological protection. According to Baller-Tech, they plan to use AI to do more things in the future. For example, referring to the application of AI technology in the chemical industry, to establish an ecological monitoring system, to incorporate more ecological and environmental factors into the entire AI computing system, and to expand Scope of species protection.
In addition, they are also exploring voiceprint recognition technology, which accounts for a high proportion of the calls of birds and beasts, adding voiceprint recognition to ecological environmental protection, breaking the limitations of a single vision, and enabling technology-enabled environmental protection work made easier and more results achieved.
Although there is no clear data to show that, the use of visual AI for wildlife protection, especially the rescue of endangered animals, is definitely a minority application. The Altay Regional Nature Conservation Association took the first step and won the first prize in the Light Challenge, which attracted more attention to beaver protection and opened up the application perspective of visual AI. Leifeng.com
What else can AI do in wildlife conservation?
This may not only be a question for the Altay Regional Nature Conservation Association to think about, but also a question for every researcher who is committed to AI for good. If AI is “conscious”, let AI warm every life on earth!
This article is reproduced from:https://www.leiphone.com/category/ai/kdRPmEuXMwtl9K0k.html
This site is for inclusion only, and the copyright belongs to the original author.