Automatic Modulation Recognition (AMR) is a critical signal processing technology in communication systems, used to identify the modulation scheme of received signals. In recent years, the rapid development of Deep Learning (DL) has made it a significant research direction in modulation recognition. This paper addresses the issue of low recognition rates in AMR due to the insufficient utilization of the temporal information of raw signals, by proposing a signal pattern recognition algorithm based on frequency domain denoising and Temporal Convolutional Networks (TCN). In our experiments, we used the standard dataset RML2016.10a and introduced a Frequency Domain Denoising Module (FDDM) to effectively mitigate the impact of environmental noise. After denoising, the in-phase/quadrature (I/Q) components of the signal are converted into amplitude/phase (A/P) components. Subsequently, the signals are vector normalized to ensure that the values of each point in the matrix are within the range of 0 to 1, thereby enhancing system stability. Finally, the preprocessed signals are fed into the TCN network for classification and recognition. Experimental results demonstrate that the proposed algorithm significantly outperforms other comparative models when handling complex modulation schemes such as 16QAM and 64QAM. Additionally, the algorithm was validated using I/Q data collected from BPSK, QPSK, and 16QAM signals with a Universal Software Radio Peripheral (USRP). The experiments indicate that the algorithm exhibits good robustness under Additive White Gaussian Noise (AWGN) channel conditions, demonstrating considerable practical applicability in engineering.