At present, my country uses face recognition technology in a brand new field-monkey face recognition, and the Northwest University research team conducts research on this technology through the field of artificial intelligence. At present, this technology is still in the experimental stage, but after capturing the facial features of golden monkeys in the Qinling Mountains, it can accurately identify 200 golden monkeys. In the future, it may be possible to accurately recognize thousands of golden monkeys in the Qinling Mountains, and the accuracy rate may reach 94%. At least from the current test results, most golden monkeys can be accurately identified.
It is learned that the scientific research team of Northwest University relies on artificial intelligence and other new technologies to develop "monkey face recognition technology", which is used to identify nearly a thousand Sichuan golden monkeys in Qinling and other places, and supports the research and protection of animals. At present, the system has been verified for applicability in 41 representative primate species and 4 carnivore groups, with an average recognition accuracy of 94.1%.
The Sichuan golden snub-nosed monkey is a rare and endangered primate species endemic to China. It lives in the Qinling Mountains of Shaanxi, western Sichuan, southern Gansu, and Shennongjia, Hubei. In order to protect and research it, the Northwestern University golden monkey research team has carried out long-term field tracking research. After difficult exploration 20 years ago, it achieved close observation and individual identification of wild golden monkeys. However, how to accurately and quickly identify individual wild golden monkeys and carry out protection accordingly has always been a problem that zoologists all over the world yearn for but cannot break through.
In response to this frontier issue, Northwestern University professor Guo Songtao and computer science experts established an animal AI research team. Using the principle of neural network, the team developed the golden monkey individual identification system based on Tri-AI technology for the first time, which realized the accurate identification of wild individuals and continuous tracking and sampling functions.
The Tri-AI system can not only be applied to different species in multiple taxa, but also can achieve continuous and unobstructed observation at night. At present, the system has been verified for applicability in 41 representative primate species and 4 carnivore groups, with an average recognition accuracy of 94.1%.
Accurate identification can help animal protection, breeding and research
Compared with Face recognition algorithm, how difficult is monkey face recognition? How will the system be more widely used in the future? In response, a reporter from the Beijing News told Guo Songtao, a professor at Northwestern University’s School of Life Sciences and deputy director of the Key Laboratory of Rare and Endangered Animals in Shaanxi Province, as well as team members Xu Pengfei, an associate professor at the School of Information Science and Technology, Northwestern University, and He Gang, an associate professor at Northwestern University’s School of Life Sciences. Conduct an interview.
Q: What are the application scenarios for individual animal identification?
A : Individual species identification can be applied to wild animal population monitoring, animal rescue, breeding and breeding of ex-situ conservation animals such as zoos, daily management of animals in animal husbandry farms, and search for lost pet animals. There are a wide range of practical needs.
For example, there are many animals on the farm. If a sheep is sick, it will be more difficult to return to the flock after treatment, and then to find it and repeat the inspection. The traditional method is to tag the sheep and wear a number or letter mark on the sheep’s ear, but this is a damage mark. The ear tags are exposed to the wind and sun, and sometimes they cannot be seen clearly, and there may be errors. Through system identification, the sheep can be found immediately and various information about it can be understood. With the identification system, special purpose animals do not need tattoos or chip implants.
At present, individual identification of birds, reptiles, and insects is difficult. We prefer to select large mammals for identification and research, such as golden monkeys, giant pandas, tigers and other rare mammals, which are expected to be expanded to other animal groups in the future.
Q: How did the researchers think about developing this system?
A: The golden monkey research team of Northwestern University has been conducting animal tracking research for a long time. The traditional method is to identify individual animals through accumulation of human experience. We observe closely and can see the clear facial features of golden monkeys through a telescope. For example, some golden monkeys have a mole under their mouth, and some have their ears torn off due to a fight. These are all natural signs. Sometimes in order to observe a specific object, we also make some artificial markings, such as colored dyes. For a period of time, as long as the location of the mark is clearly remembered, we will not lose the individual animal we observe, and we will be able to recognize it next time we see it.
But this method is not applicable in large-scale field surveys and monitoring. When doing wildlife protection, it is impossible for us to mark thousands of golden monkeys. Sometimes we use DNA identification, such as obtaining animal hair and feces, and we can detect it next time we encounter it, but this requires identification in the laboratory, which is also very troublesome and does not meet the requirements of real-time identification. Therefore, we are exploring new identification technologies according to the needs of research and protection work, and we started to tackle this technology four or five years ago.
Q: How difficult is it to develop a monkey face recognition system?
A: First of all, it does not depend on species identification technology, and requires a universal identification site. Defining the face as a universal recognition part was also decided after in-depth discussions by experts in the fields of animal ecology and computer science. The formulation of this strategy is particularly critical, as it is a difficult point facing the beginning of research and development.
AI technology requires a large amount of classified data. However, because wild animals are uncontrollable and dynamic, they will not cooperate with photographs, so it is very difficult for us to collect data in the wild. This research needs to take a lot of image data and train algorithms and models.
Animals have hidden instincts in the wild, and their hair colors may blend with the environment. For example, the Qinling Mountains in winter are gray-yellow, and the hair of golden monkeys is also this color. Under artificial conditions, it is not easy to separate individual animals from the color of the environment.
In addition, the golden monkey's facial skin has many hairy areas, the hair may have fluffy changes, and the texture features are more complex, which puts forward higher requirements on the deep learning capabilities and algorithms of the recognition system.
Q: How does the system distinguish between similar-looking monkey faces? When monkeys grow up and their faces change, can the system still recognize them?
A: At the beginning of training the model, about 20 individual golden monkeys are needed. This information must be very accurate. Each golden monkey needs to take photos of different angles to complete the system model. When a new golden monkey individual reappears, the system will extract features and compare existing information to identify the new individual. Similar to the face recognition logic, the system will calculate the recognition features of the monkey face, and carry out information marking and body system construction. Unlike people slowly accumulating experience and identifying, the algorithm can efficiently complete feature capture, comparison and identification.
After the monkey reaches adulthood, there will be no obvious changes in the face. However, there will be changes between childhood and adulthood. If the machine takes a picture of the monkey at the age of 2 and 5, the system may mistake the 5-year-old individual as a new individual. But for about 200 golden monkeys that we have studied for a long time, through continuous shooting and recording, artificial intelligence can also learn by itself, adding new characteristics to individual golden monkeys that have undergone minor changes.
It is worth mentioning that the system can also realize continuous and unobstructed observation at night. The photos and videos taken at night with the infrared camera lose the color information and show gray scale, but in this case, the system can still distinguish between individuals, indicating that the system calculates and learns to recognize feature information is extensive, including but not dependent For color information. We are still exploring the underlying principles.
Q: What other animals can the system recognize? What is the accuracy rate?
A: The system has been verified for applicability in 41 representative primate species and 4 carnivore groups, with an average recognition accuracy of 94.1%. The four groups of carnivores are tigers, lions, meerkats and red pandas. The 41 representative species of primates include Sichuan golden monkey, baboon, chimpanzee, gibbon, macaque, mandrill, etc.
System recognition is based on graphic images and facial features and requires learning a lot of data. Due to the large amount of golden monkey data, the system algorithm training has been very complete, and the recognition rate is approaching 100%, which can be used as a demonstration. However, some species are limited by the number of photographable populations, sampling conditions and other reasons, and there are not many data such as photos, so the average recognition accuracy is 94.1%. In the future, as data continues to be enriched, the recognition accuracy will increase.
At the same time, mammals such as rodents and marsupials are also potential recognizing animals of the system. In the future, this work can be extended to a wide range of industries and applications. We hope to identify the different application scenarios and needs of animals in combination with wild and captive conditions, and carry out the research and development of personalized identification functions, and realize the refined management of animal protection, breeding, breeding and research through accurate animal identification.
Face recognition technology is a kind of biometric recognition technology based on human facial features, usually using cameras or videos and images containing human faces for recognition, which can also be called facial recognition. According to the understanding, the first research was carried out in the 1960s. After the 1990s, the United States, Germany, and Japan have already realized the technology of face recognition. In addition, face recognition is mainly composed of four parts: face image acquisition and testing, face image preprocessing, face image feature extraction, matching and recognition, and so on. Of course, face recognition also includes a variety of algorithms. It can be said that in the current life, face recognition technology has gradually been applied in many fields.
In addition, according to media reports, my country currently uses china face recognition technology in a brand new field-monkey face recognition, and the Northwest University research team conducts research on this technology through the field of artificial intelligence. At present, this technology is still in the experimental stage, but after capturing the facial features of golden monkeys in the Qinling Mountains, it can accurately identify 200 golden monkeys. In the future, it may be possible to accurately recognize thousands of golden monkeys in the Qinling Mountains, and the accuracy rate may reach 94%. At least from the current test results, most golden monkeys can be accurately identified.
The reason why the accuracy rate cannot reach the accuracy of face recognition is that the monkey's face not only has a lot of hair, but also has more complex texture features, which also puts forward higher requirements for the recognition system. For this reason, it can be said that if my country's monkey face recognition technology can gradually mature, it will be a brand new technological innovation for countries around the world. Perhaps this technology can once again reach the world-leading level. After all, our country has made great achievements in various fields over the years. It can be said that for China, being able to truly master this technology is the top priority.
We must know that China once had a huge gap with Western countries in the field of science and technology. Fortunately, through years of hard work, the gap with developed countries has been gradually narrowed in many areas. For example, in the aerospace field, satellite positioning systems, and so on. It can be said that for China, the only thing to do now is to continuously innovate in order to continuously improve my country's comprehensive national strength.
In addition, although the monkey face recognition technology launched by my country is still in further trials, it is not easy for China to move from the previous imitation to the current innovation. And for China, it is inseparable from the vast number of scientists to have such achievements as it is today.