Neural network test sufficiency evaluation method based on element decomposition

The invention discloses a deep neural network test sufficiency evaluation method based on element decomposition. The method mainly comprises the steps of test element decomposition, test parameter extraction, importance clustering, mutation testing, index calculation and index evaluation. According to the method, a test sufficiency evaluation mechanism is perfected by decomposing elements of a black box test method and a white box test method and fusing the two methods, and meanwhile, a model visualization method is combined, so that neural network decision logic is more intuitive, and the interpretability of evaluation is enhanced. According to the method, the test sufficiency of the deep neural network can be effectively evaluated, and the development of a traction support artificial intelligence technology is facilitated. 一种基于要素分解的深度神经网络测试充分性评估方法，主要步骤包括：测试要素分解，测试参数提取，重要性聚类，突变测试，指标计算，指标评价。该方法通过对黑盒测试与白盒测试方法的要素分解，融合两类方法，完善测试充分性评价机制，同时，结合模型可视化方法，使神经网络决策逻辑更加直观，加强评价的可解释性。该方法能够实现对深度神经网络测试充分性的有效评价，有利于牵引支撑人工智能技术的发展