A portable microwave radar array based material classification using SpectrogramFace recognition and GoogLeNet

In recent years, radar sensor technology has garnered a lot of attention and become increasingly popular in a wide range of applications. Leveraging the powerful learning abilities of Machine Learning techniques, radar-based material classification has emerged as an area of keen interest. In our study, we employ machine learning algorithms to categorize materials based on reflected radar signals. Utilizing a portable cm-wave radar sensor, we record signals reflected by different materials and employ time-frequency analysis methods to get the spectrogram of the signals. These visual representation were then fed into a Deep Convolutional Neural Network (CNN) for classification. To simplify the process and improve efficiency, we used transfer learning with the pre-trained GoogLeNet model, avoiding the need to develop a new neural network from scratch. A key focus of our investigation is on evaluating the performance of different time-frequency analysis methods to identify the most effective spectrogram representations for material classification. Remarkably, we achieved the highest accuracy of up to 97%, which underscores the efficacy of integrating spectrogram analysis with GoogLeNet in radar-based material classification.