|Paper title||Fusion of classification approaches for landfill detection in Ukraine|
|Form of presentation||Poster|
Based on the latest satellite data from the European Space Agency under the Copernicus program, which has been available since 2016, it is possible to receive satellite images of every land cover point on a regular basis. Such data are provided with a spatial resolution of 10 meters, which opens up a huge number of open problems for scientists and researchers.
At the same time, there is an urgent problem for Ukraine that almost every country is currently struggling with - the monitoring of waste storage, including human activities, and their storage or use for re-production or disposal. About 22,6 thousand unlicensed disposal dumps were located according to information of Ministry for Communities and Territories Development of Ukraine. It happens because a lot of legal landfills are filled and for some territories, they aren’t existing at all. That is why there is a need for monitoring and tracking of existing landfills. Currently, there are services that track the location of landfills, but these are only point indicators. Such services do not provide information on the area of the landfill and its changes over time, while our landfill monitoring algorithm, thanks to the possibility of using historical satellite data, makes it possible to track the area of landfills over time.
There are a lot of studies that describes different technologies of landfills monitoring based on satellite data. Each study based on different satellite data and artificial intelligence approaches. For example, for Iran authors compare four satellite providers with different parameters and spatial resolution for mine waste dump monitoring , and in  authors analyzed different indexes and temperature of landfills. Different CNN architects are used as neural network algorithms in the world ,  for landfills identification.
In this paper, within the project “Landfill detection and monitoring service” we propose an algorithm based on neural networks and satellite data, which will allow automated monitoring of landfills on remote territory, as well as the ability to assess retrospective information and monitor changes in time. During the project implementation there were some difficulties. In particular, as landfills are quite dynamic and often change their contours, the neural network model had to be set up on the basis of one satellite image for a specific date and accordingly the time series of satellite data was used only to track the dynamics of a particular landfill.
Each of the classification methods has its advantages and disadvantages. The pixel-based method identifies artificial objects well, but does not cope well with the separation of landfills from artificial objects and quarries or sands. The object-based method, in turn, identifies landfills well, but also identifies some parts of cities that are similar in spectral characteristics with landfills. In our study the main problem is the separation of landfills from quarries and artificial objects was solved. In this, we used a fusion of pixel-based  and object-based  classification, which helped to identify those areas that belong to the class of landfills.
For the pilot areas (Olhynska, Pokrovska, Myrnohradska and Kurakhivska regional territorial communities of Donetsk region, Ukraine) the first results have already been obtained, in particular the landfills found according to Planet and Sentinel-2 satellites. In the future, we plan to expand our product first of all for the Donetsk region, and then to the whole of Ukraine. We plan to involve the city and municipalities, environmental authorities, directly - the Department of Ecology and Natural Resources of the Donetsk State Regional Administration and the State Ecological Inspectorate in the Donetsk region. The preliminary results are presented on the web interface at the link http://inform.ikd.kiev.ua/ldms/.
The project “Landfill detection and monitoring service”, which is the winner of the EastCode2021 national innovation competition, is implemented by the NGO “Open Initiatives” under the technical administration of Center42 within the UN Peacebuilding and Peacebuilding Program and with financial support from the governments of Denmark, Switzerland and Sweden.
. Khosravi, Vahid, et al. "Satellite Imagery for Monitoring and Mapping Soil Chromium Pollution in a Mine Waste Dump." Remote Sensing 13.7 (2021): 1277.
. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” Iclr, (2015) 1- 14.
. Adedeji, Olugboja, and Zenghui Wang. "Intelligent waste classification system using deep learning convolutional neural network." Procedia Manufacturing 35 (2019): 607-612.
. Torres, Rocio Nahime, and Piero Fraternali. "Learning to Identify Illegal Landfills through Scene Classification in Aerial Images." Remote Sensing 13.22 (2021): 4520.
 Kussul, Nataliia, Mykola Lavreniuk, and Leonid Shumilo. "Deep Recurrent Neural Network for Crop Classification Task Based on Sentinel-1 and Sentinel-2 Imagery." IGARSS 2020-2020 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2020.
. Shumilo, Leonid, Nataliia Kussul, and Mykola Lavreniuk. "U-Net Model for Logging Detection Based on the Sentinel-1 and Sentinel-2 Data." 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. IEEE, 2021.