中圖分類號：TP393.08文獻標識碼：ADOI:10.19358/j.issn.2097-1788.2025.06.001
引用格式：周品希，沈岳，李偉. 基于深度學(xué)習(xí)的物聯(lián)網(wǎng)入侵檢測系統(tǒng)綜述［J］.網(wǎng)絡(luò)安全與數(shù)據(jù)治理，2025，44（6）：1-10.

A review of IoT intrusion detection systems based on deep learning

Abstract： While the interconnection of smart devices in the Internet of Things promotes social progress, it also leads to increasingly complex security threats due to device heterogeneity, protocol diversity and resource constraints. Traditional intrusion detection systems rely on feature matching and rule definition, and show limitations when facing new attacks and dynamic attack patterns. This paper systematically sorts out the application progress of deep learning technology in the intrusion detection system of the Internet of Things. Through comparative analysis, it is found that the model based on deep learning is superior to traditional methods in detection accuracy and real-time performance, and has outstanding performance in processing spatial features and capturing temporal dependencies. Unsupervised learning and integration methods effectively improve the detection robustness in small sample scenarios by generating adversarial samples and integrating the advantages of multiple models. Current research still faces challenges such as high data annotation costs, limited edge computing resources, and insufficient adaptability to dynamic attacks. This paper summarizes and discusses the directions that future research should focus on, such as lightweight and cross-modal data fusion, to provide theoretical support for building an efficient and adaptive Internet of Things security protection system.

Key words : network security; Internet of Things; intrusion detection; deep learning