Enhancing Soil Carbon Content Classification Using Transfer Learning and FrFT on Millimeter Wave Radar Array Signals with Varied Backgrounds

The precise quantification of soil carbon content is crucial for understanding soil health and its impact on climate change. This study investigates the classification of soil carbon content using millimeter wave radar array signals, focusing on a comparative analysis of Support Vector Machines (SVM), Continuous Wavelet Transform (CWT), and Fractional Fourier Transform (FrFT) methods. Furthermore, raw time and CWT-processed signals using SVM are used as baseline to establish the fundamental classification performance. The key contribution of this research is the innovative use of FrFT, where the transformed signals are converted into images and subsequently fed into GoogLeNet through transfer learning. This approach leverages the powerful feature extraction capabilities of GoogleNet to enhance classification accuracy. Experiments were conducted on a dataset comprising millimeter wave signals reflecting varied backgrounds with the primary focus on one specific distance measurement. Results provided show that high resolution millimeter wave signals achieved up to 100% validation accuracy at multiple fractional orders. The improved accuracy highlights the potential of integrating advanced signal transformation methods with deep learning models for soil carbon content classification.