RUDN mathematician improved the machine learning algorithm for recognizing images from satellites

RUDN mathematician improved the machine learning algorithm for recognizing images from satellites

Mathematician from RUDN proposed a machine learning method that allows to automatically recognize images from satellites and aircraft radars. The algorithm can determine which type of plant is planted in the fields more accurately than previously existing developments.

Modern satellite and radar systems can be used to automatically monitor earthquakes, volcanic eruptions, fires, and other disasters, as well as to monitor the condition of soil, vegetation, and rivers. To automate this process, complex algorithms need to recognize and classify objects, allowing the computer to understand from a set of pixels what is depicted on the image. Machine learning is used for this purpose, a computer “looks through” thousands of examples and thus learns to recognize images independently. To improve machine learning results, a combination of several training algorithms is often used. This yields more accurate solutions than each of them separately. The RUDN mathematician developed such an ensemble method using three algorithms to process data from multiple sources.

Mathematicians used data from five RapidEye mini-satellites and the UAVSAR airborne radar on July 5 and 7, 2012, they captured the same area in Canada. The RapidEye imagery was acquired in five bands of the light spectrum: blue (B), green (G), red ®, near-infrared (NIR) and a region called “red edge” (RE) where the reflection of green vegetation is dramatically enhanced. The data contained 38 features: spectral channels, vegetation indices, texture parameters, and etc. Their spatial resolution, that is, the minimum object size distinguishable in the images was about five meters. UAVSAR radar images included 49 different features, with a spatial resolution of about 15 meters. Mathematicians compared the images with reference data on the area collected in the summer of 2012. They identified seven types of plants, broadleaf plants, rapeseed, corn, oats, peas, soybeans and wheat. The new algorithm was “trained” based on examples of images and planting type data, and then compared its prediction with the results of other programs based on a similar principle.

The new method showed higher accuracy in interpreting images, both on large and limited amounts of examples for training algorithms. If training was performed on 5% of all data, the new algorithm recognized images correctly at least 65% of the time, while the other algorithms were correct in 52-60% of cases. With an increase in the share of training data to 50% of the total volume, the accuracy of the new algorithm increased to almost 90%, and the other algorithms increased to 75-86%. Thus, the application of the new algorithm was found to be more effective.

“Our method can be proposed for a land use and land cover classification system using data from different sources. For example, Landsat or Sentinel constellation satellites,” Vladimir Razumny, Ph.D., Associate Professor at the RUDN Mechanical and Mechatronics Engineering department.

The article was published in International Journal of Image and Data Fusion

Scientific Conferences View all
03 Nov 2017
RUDN University organized the first 5G Summit R&D Russia on June 19 - 20, 2017
Main Publications View all
15 Nov 2017
RUDN University scientists publish results of their scientific researches in highly-recognized in whole world and indexed in international databases journals (Web of Science, Scopus ect.). That, of course, corresponds to the high status of the University and its international recognition. Publications of June-September 2017 ( In Journals of categories Q1-Q3)
Similar newsletter View all
26 Dec 2022
Tissue architecture, cell organization, biomedical products: RUDN University opens a new research and educational resource center

On October 4, the Research and Educational Resource Center (REC) of innovative technologies of immunophenotyping, digital spatial profiling and ultrastructural analysis (molecular morphology) opened at the RUDN.

26 Dec 2022
RUDN scientists suggested how to help the soils of Zaryadye Park

RUDN University scientists conducted a comprehensive soil and environmental survey and took more than 80 soil samples in Zaryadye Park. An assessment of the physicochemical, microbiological, and ecotoxicological properties of soils made it possible to develop recommendations and a plan for the care of soils in analogous landscapes in the park.

26 Dec 2022
RUDN University Chemist Creates Nanofilter to Clean Water from Toxic Dyes

RUDN University chemist with colleagues from India and Korea created a nanofilter for water purification from synthetic dyes. The graphene-based composite can quickly remove up to 100% of harmful compounds from water, and it can be used up to seven times without losing efficiency. In addition, the synthesis of the nanofilter itself is economical and environmentally friendly.

Similar newsletter View all