二进制代码漏洞检测是一个重要的研究问题。针对当前基于静态分析的二进制代码漏洞检测系统普遍存在误报率高,检测粒度粗,且依赖专家经验等问题,提出了用库/API函数调用程序切片细粒度表示二进制程序,并引入深度学习技术,自动检测二进制程序中库/API函数调用相关漏洞,设计并实现了一个基于深度学习的二进制代码漏洞检测系统——BVDetector。该系统通过对二进制程序进行控制流和数据流分析,从程序中提取库/API函数调用的程序切片,并为程序切片添加有无漏洞的标签;然后将程序切片转换成符合漏洞检测深度学习模型输入的向量表示;最后用程序切片的向量表示及其标签训练和优化深度学习神经网络模型。相关实验结果显示,BVDetector能够有效检测出二进制程序中库/API函数调用相关漏洞,并且将二进制漏洞检测的误报率降低至2.3%,漏报率降低至8.2%。(Binary code vulnerability detection is an important research issue. Aiming at the problems of high false alarm rate, coarse detection granularity and dependence on expert experience in current binary code vulnerability detection system based on static analysis, this paper proposes to use library/API function call program slice to represent binary program with fine granularity, and introduces deep learning technology to automatically detect the related vulnerabilities in library/API function call of binary program. A deep learning-based method is designed and implemented. Binary code vulnerability detection system - BVDetector. By analyzing the control flow and data flow of binary program, the system extracts the program slices of library/API function calls from the program, and adds vulnerability labels to the program slices.)