RAPSIM

Target Classification

Various features extracted from measurement and/or synthetic range profiles and micro-Doppler data of radar, sonar, or optical sensors, along with different classification structures (statistical, deep learning, etc.), are utilized to perform classification activities for land (such as tanks, missile launchers, etc.), air (drones, helicopters, aircraft, etc.), or sea (fishermen, cargo ships, frigates, etc.) platforms. A synthetic data library providing input for target classification activities is being created using the high-frequency track estimation software (RASES) developed within TÜBİTAK.

RAPSIM Target Classification

RAPSIM Target Classification conducts classification activities for land, air, or sea platforms using measurement and/or synthetic data obtained from radar, sonar, or optical sensors. In this process, various features and different classification structures are employed to analyze the characteristic properties and behaviors of targets, allowing for the differentiation between classes of objects. Target classification plays a critical role in target detection and threat assessment in defense systems.

Target Types

In classification studies, a comprehensive list is created based on input data possibilities for the classification process of all target types that real systems may encounter. In this context, various civilian or military target types are taken into account, such as military trucks and missile launchers for land platforms, aircraft and drones for air platforms, and fishing vessels and tankers for sea platforms.

Input Data Types

In classification studies, both measured data obtained from real systems and synthetic data created through RASES software can be utilized as "input data." The input data, obtained or generated while considering the data formats created by real systems, can take various forms such as "Range Profiles," "Inverse Synthetic Aperture Radar (ISAR) Images," or "Micro-Doppler Profiles." Computer-aided design (CAD) models of targets are employed in the creation stage of synthetic input data.

Feature Types

Appropriate features are extracted and provided to the classifier by considering the characteristics of the data created for the platforms under classification studies and the related platforms. In this context, a wide variety of features are obtained and used in basic spatial (length, symmetry, etc.), statistical (mean, standard deviation, etc.), or transformation coefficients (wavelet, Fourier-Mellin, Gabor, etc.) forms, considering the fundamental characteristics of the data.

Classifier Types

In classification studies, a variety of classifier structures can be applied to achieve the highest recognition performance, either individually or on an average basis. These structures include basic Bayesian, proximity-based (k-NN), cost function optimization (perceptron, least squares, support vector machines, artificial neural networks), or deep learning (convolutional neural networks) approaches.