In magnetic flux leakage (MFL) detection of coiled tubing, deep learning techniques have gained widespread attention due to their efficiency and automation in defect identification. However, training deep learning models typically requires large volumes of high-quality labeled data. In practice, the high cost of data acquisition and the scarcity of defect samples hinder data collection, leading to overfitting and reduced generalization. To address this challenge, this paper proposes a Frequency-Aware Wasserstein Generative Adversarial Network with Gradient Penalty (FA-WGAN-GP) for high-quality MFL signal generation under limited sample conditions. The proposed model introduces a spectral attention mechanism in the generator to guide the network in learning key frequency-domain features of defect signals. Additionally, a dual-domain discriminator structure is designed to evaluate the authenticity of generated signals in both time and frequency domains, enhancing training stability and physical consistency. Comparative experiments on five typical defect types demonstrate that the proposed method outperforms traditional GAN-based approaches in signal fidelity, spectral structure reconstruction, and error metrics. The proposed approach offers practical value in addressing data scarcity and improving model robustness and generalization.