Abstract: Aiming at the existing problem that the correlation between malicious URL characters is not obvious and the existing methods cannot keep up with the speed of malicious URL updates and changes, we propose a malicious URL detection method that combines OMNI-dimensional dynamic convolution (ODConv) and DPCNN model. It used WordPiece to URLs for character encode, and semantic features between characters in URLs were preserved, and long-distance dependent text relationships were extracted using a full-dimensional dynamic attention mechanism. The final output were classification detection results. The experimental results show that the proposed method has improved accuracy and F1 value in both binary and four-class classification tasks compared with existing methods for malicious URL detection, and can detect malicious URLs more accurately.