RADAR KESİT ALAN ANALİZİ İÇİN YAKIN ALANDAN UZAK ALANA MAKİNE ÖĞRENMESİ TABANLI DÖNÜŞÜM ALGORİTMASI GELİŞTİRİLMESİ

Abstract

Radar systems inherently operate in the far-field region of electromagnetic waves. Measurements with radars are conducted in either indoor or outdoor environments in the far-field region. While outdoor measurements offer advantages such as a wide observation area and comprehensive data collection, they also come with disadvantages due to environmental conditions, weather, and other external factors that can introduce unwanted effects on radar signals. To avoid this uncontrolled situation, measurements are performed in indoor environments. Conducting measurements indoors allows for the control of external factors, making the measurement process more predictable. However, the dimensions of anechoic chambers designed for indoor measurements typically do not cover far-field distances. As a result, measurements need to be performed in the near-field and then transformed into the far-field, which often involves a lengthy and costly conversion process. As a solution to this problem, the aim is to make this transformation process more efficient and effective by using machine learning algorithms in the conversion of radar measurements from the near-field to the far-field. The data comprising of a total of three million scattered electric field information from targets illuminated by electromagnetic waves, covering the X-band (8-12 GHz) in different angles and frequencies, has been collected through the ALTAIR FEKO program. These data were trained with regression algorithms available in MATLAB application and iv compared with far-field data. The machine learning algorithm was trained to predict far-field data using near-field data. The predicted data were processed with MATLAB to create images of Inverse Synthetic Aperture Radar (ISAR). Subsequently, scattering centers were calculated with the predicted data, and ISAR images were reconstructed with the calculated scattering centers and compared with the original images. The impact of unwanted noise signals arising from the laboratory or measurement equipment on the machine learning algorithm was examined, and the results in the ISAR images were compared. The machine learning algorithm successfully generated ISAR images for both noisy and noiseless signals, demonstrating that more effective results could be obtained with an increase in training data.