Source: ATYUN AI platform
In studying biodiversity of a region or a specific monitoring of wild animal species, one of the most effective tools used by scientists and conservationists is the camera trap.
Camera trap refers to the use of motion sensors, infrared detectors or other beam as a remote camera trigger organ. It is often used to shoot the photographer is not easy to photograph the screen directly, use more research in the field of ecology, such as supervision of hunting, wildlife observation, looking for rare species.
When an animal enters the range of the camera traps, its sensor will detect its presence and automatically trigger the camera shutter release to capture pictures of animals (or series of pictures).
Zoological Society of London (ZSL) extensive use of camera traps, the organization is perhaps the most famous London Zoo organization, but it itself is an international scientific, conservation and educational charity.
Although this technique for monitoring pygmy hippo communities Sumatran tigers in Indonesia and Liberia are very valuable, but it also has negative aspects, ZSL protection technical director Sophie Maxwell said:
"Project to deploy a large number of camera traps in a few months may be sent back millions of images. In the past, scientists would sit there to manually check each image, and then create a species identification data required for each report on biodiversity."
She said, basically, even a medium-sized project, also takes about nine months to complete the report. In this project, because the hunters poaching or intrusion, in any case, the situation may change. This may mean that the region's animal protection policy needs to be changed in accordance with the updated information.
Speed up the process
Over the years, ZSL has been exploring how to use technology to speed up data processing. For example, by recruiting "citizen scientists" to help manually check image. But now it seems the reason, artificial intelligence, machine learning, and more specifically the completion of the work might improve the accuracy and greater impact, this is the ZSL cooperation with Google.
Since the beginning of 2017, ZSL is has been one of the few organizations with the technology giant, and Urban Outfitters fashion companies, and media and entertainment giant Walt Disney together, improve its Cloud AutoML Vision tool.
Google Cloud AutoML Vision is the first service Cloud AutoML services provided, designed to help companies with limited expertise in machine learning machine learning. For example, AutoML Vision focused on helping them build a custom machine learning model, using a drag and drop interface to upload images, training and management model, and then deploy them directly to the Google Cloud.
In the ZSL, the technology has been proven to be very suitable, Maxwell said:
"In the past, artificial intelligence and machine learning technologies for organizations like ours it is difficult to enter. You need a data scientist, write your own model, and the commercial pre-trained models tend to be very basic, but it can not provide we need a complex level of detail, such as identifying a population of a particular species. "
"So we've been on availability with Google to do AutoML Vision tools. We are trying to build a model based on our current data, we often do camera trap places, such as Borneo (Borneo) and Costa Rica (Costa RIca), we It can be prepared when new data enters a good model, so we can this year compared with last year's data. over time, we will provide specific locations, especially in specific environments, such as forests, grasslands or Antarctica, build customer model. So far, the results look pretty good. "
Cloud-hosted model ML
She said it was an exciting job because conservation technical team consists of eight members, they do not have specific data science skills, but to build such a model, it will be hosted in Google Cloud, and the it is open to other environmental organizations around the world.
In addition, you can also call these cloud-based model from within the application via the API. For wild animals and for monitoring threats, ZSL real-time detection technology has occurred, it relies on sensors and camera trap connected to track animals, poachers determine the intrusion and alerts the administrator may need to respond to the situation.
But this is just the beginning, Maxwell said:
"Our long-term goal is to provide health checks for the Earth, which is obviously a lot of ambition, but as the start of these studies and the results more quickly recover, our views on biodiversity will grow."
She said that, in a sense, technology trends is one side of the ZSL. Sensors become cheaper, cloud-based processing and storage capacity is almost limitless. Companies like Google is clearly committed to machine learning technology spread to new areas. Maxwell concluded:
"But we are a charity, so we rely on partnerships and funds, and always welcome to help. With more support, we can do the work would be amazing."
This switched ATYUN artificial intelligence media platforms, the original link: Zoological Society of London Wildlife image identification data are collaborating to develop models with Google, you can be taken from the wild in
more recommendations
Sagitar get-Year 2019 Stevie Award
AI radar data alone can be measured to classify objects on the road
Clarifai: What is the depth of learning, and why is it important?
