The rapid expansion and global reach of social media have made it a cornerstone of information dissemination. However, the escalating spread of fake news presents a pressing global issue. Fake news distorts public understanding, manipulates emotions, and jeopardizes social stability. Current models primarily focus on textual content or explicit image features, overlooking the latent semantics in image frequency-domain information. Additionally, social media datasets often suffer from limited visual data or text-image discrepancies, complicating detection efforts. To tackle these issues, we propose a streamlined approach comprising: 1) a Frequency-Domain Detection Module that applies discrete cosine transform to extract image frequency-domain features; and 2) an Image-Augmentation and Cross-Modal Consistency Module that enriches visual data through augmentation, employs pretrained vision models for feature extraction, and aligns text-image semantics via consistency learning. Experimental results show enhanced accuracy, confirming the method’s efficacy.