As deepfake technology evolves, it lowers the barrier for tampering with images and videos, making such manipulations more accessible to those with malicious intent. In response, numerous deepfake detection methodologies have been developed. Nevertheless, as deepfake technologies advance, these detection methods often struggle to maintain accuracy when faced with new, previously unseen forgery techniques. Recent research has highlighted that wavelet analysis offers unique advantages in identifying subtle manipulation traces that are not easily discernible in the spatial domain. This approach effectively captures fine-grained and holistic texture details in manipulated faces. However, current methods relying on wavelet analysis tend to overfit on the identity features of faces during the feature extraction process, failing to fully leverage distinctive texture features. Consequently, this limits their feature representation capabilities and overall model generalizability. To tackle the limitations of existing approaches in adapting to emerging forgery techniques, we propose an innovative deepfake detection method. Our solution, inspired by the principles of wavelet transform, introduces a plug-and-play module known as WCAT, which is designed to generate multi-scale attention feature maps. By combining wavelet attention with spatial features, our method enhances both the accuracy and the generalization capability of deepfake detection models.