【摘要】 近年来，模糊测试技术已经在软件系统漏洞挖掘和脆弱性检测中得到了广泛的使用。模糊测试主要对收集到的种子进行特定变异来获得大量的测试用例，用于执行目标程序，触发程序中潜在的崩溃。在测试过程中，如何从种子队列中选择合适的种子进行变异是影响模糊测试性能的一个重要因素，目前的策略存在一定的盲目性，影响了模糊测试的效率。因此，本文提出了一种基于深度学习测试用例的输入文件分类方法，利用深度学习技术对模糊测试过程中生成的大量测试用例和执行覆盖率情况进行模型训练，学习输入文件的结构和覆盖状态之间的关系。本文基于AFL实现了一种利用此方法判断种子优先级的模糊测试策略，经过实验评估，该方法对覆盖率提升和崩溃的发现有一定的积极作用。更多还原

【Abstract】 In recent years, fuzz has been widely used in software vulnerability mining and vulnerability detection. Fuzz focuses on specific variations of collected seeds to obtain a large number of test cases that can be used to execute the target program and trigger potential crashes in the program. In the testing process, how to select the right seed from the seed queue for mutation is an important factor affecting the performance of fuzzy testing. Therefore, this paper proposes a fuzz method based on deep learning.Deep learning is used to conduct model training on a large number of test cases generated in the fuzz and the execution coverage,and learn the relationship between the structure of the input file and the coverage. Based on AFL, this paper implements a fuzz tool for seed selection using deep learning model. After experimental evaluation, this tool has some improvement in coverage improvement and crash detection.更多还原