Rapid and accurate identification of unknown malware and its variants is one of the important research directions in the field of cyberspace security. Based on a brief description of the significant research value of malware detection, the existing deep learning-based malware detection techniques and methods are summarized in consideration of the current situation of domestic and foreign research. Firstly, the traditional detection techniques are sorted out from static, dynamic and hybrid detection methods respectively. Secondly, the malware classification and identification methods based on deep learning are summarized from the malware feature extraction methods based on sequence features, image visualization and data enhancement. Finally, the technical difficulties and future development trends of malware feature extraction and identification based on deep learning are analyzed and foreseen.