Classification of nonlinear targets using harmonic radar spectrogram images

Harmonic radar is a specialized type of radar system developed for the detection of nonlinear targets. Circuit components such as diodes, mixers and amplifiers are some known nonlinear devices. These targets exhibit distinctive behavior by emitting harmonic frequencies of the incident wave. This property is primarily due to their nonlinear voltage-current characteristics, which cause them to generate harmonics of the input signal. When a signal is applied to these devices, they respond by producing harmonic frequencies. Harmonic radar systems operate by processing the harmonic components present in the reflected electromagnetic waves to identify the presence of nonlinear targets. This exploitation of the nonlinear behavior of electronic circuits provides harmonic radar with a significant advantage in detecting targets in environments with substantial background clutter. Additionally, this technology enables the detection and classification of concealed objects, such as buried devices, hidden weapons, or other man-made items that produce nonlinear electromagnetic responses. This study presents a new approach to classify different nonlinear circuits using harmonic radar response. Unlike prior approaches, a Convolutional Neural Network (CNN) model is used to process spectrogram images that are generated from the recorded harmonic response data. These harmonic responses are obtained in a simulated environment. A Linear Frequency Modulated (LFM) pulse is chosen as the input signal for the simulations. Because LFM signals vary in frequency over time, they allow detailed observation of how the harmonic content of the response evolves, thus offering deeper insights into circuit behavior. A large dataset is created by simulating different noise conditions and adjusting circuit parameters that affect the nonlinear response. Spectrograms are constructed as 2D color images capturing frequency and amplitude variations over time. These images are used to train a CNN model to perform classification, resulting in an accuracy rate exceeding 80.6%. The results demonstrate that CNNs are effective in classifying spectrograms derived from harmonic responses of nonlinear circuits. Keywords: Harmonic radar, Nonlinear target, Classification, Convolutional neural network, Spectrogram