Target recognition method based on wavelet scalogram and deep neural network
TL;DR
A target recognition framework that integrates continuous wavelet transform with deep learning and substantially improves the capacity of deep neural networks for analyzing non-stationary signals and offers a practical solution for target recognition applications.
Abstract
Conventional deep neural networks face performance degradation when handling non-stationary signals due to their inability to capture time-varying patterns effectively. To overcome this limitation, this paper presents a target recognition framework that integrates continuous wavelet transform with deep learning. The raw signal is first transformed into a time-frequency scalogram representation, which preserves local spectral variations over time. This two-dimensional representation is then fed into a convolutional neural network for feature extraction and classification. Comparative evaluations demonstrate that the proposed approach enhances recognition accuracy by 15.3% relative to time-domain baseline methods. Under challenging noise conditions with a signal-to-noise ratio of 10dB, the system maintains 92.4% accuracy, confirming its robustness. This work substantially improves the capacity of deep neural networks for analyzing non-stationary signals and offers a practical solution for target recognition applications.