The era of decentralized supervision has come. Mining is the main economic activity in the Amazon region, and mining has made a significant contribution to the national economy on this land. However, mining has also brought negative impacts on the land, such as vegetation inhibition, soil destruction, river and forest degradation and fauna migration. To make matters worse, in 2021, statistics pointed out that illegal mining activities in Brazil's indigenous reserves increased by nearly 500% in 10 years. About 28% of Brazil's gold exports in 2019 and 2020 may come from illegal mines.
Author Wang Yue
Editor Chen Caixian
How to control illegal mining and effectively monitor mining, so as to protect local vegetation, soil and water resources, has become a top priority in the Amazon region.
In response to this situation, Pulitzer center and Earthrise media jointly launched Amazon mining watch. The two non-profit organizations cooperated to investigate the rampant expansion of illegal mining in Brazil and Venezuela. With this successful experience, they decided to extend their method to the whole Amazon region.
The platform is algorithm driven. Its principle is to detect gold mining and other open-pit mining activities in the Amazon rainforest by analyzing satellite images. Journalists, scientists and ordinary citizens interested in Amazon mining can use this platform for remote monitoring.
1. Everyone is a righteous policeman
In its beta version, the platform analyzes 326 million high-resolution images covering the entire Amazon region every four months, including nine countries with a size of about 6.7 million square kilometers. The algorithm has locked in an area characterized by mining activities, with an area of 68000 square kilometers, about four times the size of the whole new york city.
Enter the map option on the official website to see the bright yellow mining characteristics on this densely vegetated land:
When you zoom in, you can see the longitude and latitude of the mining area:
The local location can be known by exploring the longitude and latitude. The next work is to study whether the local area has obtained a legal mining license and whether it operates beyond the normal mining time. The main goal of Amazon mining observation platform is to encourage journalists, researchers and people from all walks of life who care about the Rainforest Ecology to use data as a springboard to investigate and truly implement the data collected by artificial intelligence into applications.
2. Principle of platform algorithm
Mine detector is a statistical calculation model of artificial neural network. Designers train the mine detector to separate the mine from other terrain areas by inputting the manual marking examples and other key features of the mine in sentinel-2 satellite images into this model.
This lightweight convolutional neural network will operate from each block of 44 x 44 pixels (440m x 440m) in the L1C data feedback of sentry 2. Each pixel can capture 12 bands of visible and infrared light reflected from the earth's surface. The designers averaged the data generated by the sentry over a period of four months (composite median) to reduce the impact of clouds, cloud shadows and other obstacles.
During operation, the artificial neural network will evaluate whether there are recent signs of mining activity in each block, and then move another 140 meters to evaluate the next area. This process is carried out in the whole area with mining activities. The network will make 326 million separate assessments, covering 6.7 million square kilometers of the Amazon basin every four months.
At present, the developers of Amazon mining observation platform have disclosed the data and code of the system at< https://amazonminingwatch.org/en/code >。 The system was developed for the Amazon region, but scholars believe it can also be used for other tropical biological communities. Developers welcome external development and use of code, but subject to the open MIT license terms, creating data sets and use models currently require access to the Cartesian laboratory platform.