RUDN engineers name best machine learning methods for processing radar data
Engineers at RUDN University compared four machine learning methods used to process radar data. The researchers named the most effective and fastest methods. The study is published in the European Journal of Remote Sensing.
Newswise – Images of the Earth’s surface and other planets are obtained using Synthetic Aperture Radar (SAR). The radar is placed on a spacecraft or a carrier aircraft. It scans the surface and simultaneously tracks its position. As a result, detailed maps of the surface are obtained and their quality does not depend on the weather or time of day. The most common type of these radars is PolSAR. Machine learning methods are used to process the radar data. Due to the differences in the algorithms, they work with different precision and speed. Therefore, with an incorrectly selected algorithm, the calculations turn out to be less precise or require more time for the calculations. RUDN engineers compared the four most popular methods and found out which one is the most effective.
âThe classification of PolSAR data is still one of the favorite topics of remote sensing researchers. A wide range of algorithms are used for this. The best known of these is the Support Vector Machine (SVM), which is widely used to classify PolSAR. However, there has been no research into the use of certain extended versions of SVM so far. We compared these methods to classify the PolSAR data, âsaid Prof. Yury Razoumny, head of the department of mechanics and mechatronics, director of the Academy of Engineering. at RUDN University.
RUDN engineers and their overseas partners compared four methods: the Support Vector Machine (SVM) and its three modifications – the Support Vector Least Squares Method (LSSVM), the Relevance Vector Machine (RVM) and the import vector machine (IVM). Their work was tested on three datasets obtained from PolSAR: images from Flevoland (Netherlands), Foulum (Denmark) and Winnipeg (Canada). The first and third datasets included large agricultural areas. Foulum’s images were mostly forests, agricultural fields, and populated areas. The machine learning algorithms had to figure out how each piece of land is used (where the wheat grows, where the forest grows, where the river flows, etc.). The algorithms were trained on 5%, 10%, 50%, and 90% of the data, and the rest were used to test their performance. The efficiency of the algorithms was evaluated by an indicator varying from 0 to 1, with the ideal classification corresponding to one, as well as the time required for learning according to the algorithm.
LSSVM was found to be the fastest – for any amount of training data and for all three districts. For example, for a Foulum with 50% of the data provided for training, LSSVM took less than 0.5 seconds and the rest of the algorithms took 12 to 15 times longer. However, SVM has proven to be the most efficient. It showed the highest learning rate for almost all data volumes for Winnipeg and Foulum: 0.78 for Foulum and 0.69 for Winnipeg. Second place in both cases was taken by IVM – 0.76 and 0.68, respectively.
âSVM was found to be more efficient, more precise and more stable when classifying two of the three datasets. Another conclusion we have drawn is the incredible speed of LSSVM compared to other methods. LSSVM produces comparable accuracy at a rate 12 times faster than SVM and about 15 times faster than RVM and IVM. Therefore, LSSVM can be considered a worthy modification of SVM with acceptable precision and higher speed, âsaid Javad Hatami Afkoueieh, PhD student at RUDN Engineering Academy.