Convolutional neural network-oriented mutation coverage test method and computer storage medium

Abstract

The invention discloses a convolutional neural network-oriented mutation coverage test method and a computer storage medium, and the method comprises the following steps: 1) setting n mutation operators, and respectively injecting the n mutation operators into a to-be-tested convolutional neural network program P to obtain a mutation program set {P1, P2, P3,and the like, Pn}; 2) training the variation program set {P1, P2, P3, and the like, Pn} by using a training data set D to obtain a variation model set {M1, M2, M3, and the like, Mn}; 3) testing the original model M and the variation model set {M1, M2, M3, and the like, Mn} by using a test data set T; and 4) comparing the test accuracy of all the models, and selecting the model with the highest accuracy. According to the invention, the defect that the traditional test method is difficult to ensure the test sufficiency of the convolutional neural network application program is solved. The test sufficiency of the convolutional neural network can be effectively improved, the method is more effective in neural network model testing, the local optimal model can be found out according to the test accuracy, and the quality and safety ofthe convolutional neural network application program are effectively guaranteed.

Description

technical field

[0001] The invention relates to a software testing method and a computer storage medium, in particular to a convolutional neural network-oriented mutation coverage testing method and a computer storage medium. Background technique

[0002] The practical application of convolutional neural networks in image classification and recognition, natural language processing and other fields has achieved great success, and many safety-critical fields are also eager to introduce convolutional neural networks. However, due to some errors in convolutional neural network systems recently, people are paying more and more attention to the security and reliability of convolutional neural network applications. Current testing methods mainly consist of white-box differential testing algorithms for systematically generating adversarial examples covering all neurons in the network. For convolutional neural network testing adequacy, existing software testing adequacy methods and ...

Images