Signal Feature Recognition in Time-Frequency Domain Using Edge Detection Algorithms

We propose a method of discerning components in multicomponent, stationary and nonstationary signals by application of edge detection techniques to the time-frequency (TF) plane. The approach is based upon the use of a robust to noise computer vision edge detection algorithms, which can be used to precisely mark the position of the component in the TF plane independent of its length, frequency or shape. The results show the proposed method correctly detects positions of stationary signals with low error even in signals heavily corrupted by Additive White Gaussian Noise (AWGN) and other color noise environments, tested for Signal-to-Noise Ratio (SNR)of OdB and 6dB. Positions of nonstationary components in the TF plane are detected with error of less than 6%. Results with synthetic signals and a real-life signal (bat-echolocation) indicate that the method can be used in identifying components in noisy environments using a computationally less costly method that outperforms previously proposed adaptive methods by offering faster computational speed and smaller processor workload. Closer to optimal detection can be achieved with a combination of edge detection operators and thresholded image segmentation procedures.