Research On Classification Method Of Hash Function Based On Machine Learning

The classification of cryptographic algorithms is of great importance for cryptanalysis research,and researchers have made some progress in this area,but they are still more concerned with the features directly reflected in the randomness test of ciphertexts,and less on further mining.In addition,although scholars have carried out some research based on machine learning,there are not many schemes that try to use various neural network models,and the multi-classification problem still needs to be extended.In addition,scholars who have studied the classification problem of cryptographic algorithms have mainly focused on block cryptographic algorithms,and less attention has been paid to the classification problem of hash functions.To address the above problems,the following work is carried out in this thesis.(1)In terms of feature extraction: Firstly,on the basis of randomness test,similarity calculation is performed for each return value of the hash value sequence generated by each hash function after randomness detection,and the three detection terms with the largest distance values are obtained,and then the new feature extraction algorithm FLLU balanced statistical detection algorithm is reconstructed by combining its detection principles.Then,the idea of convolutional neural network is used to construct a new feature extraction algorithm FCN hash feature extraction algorithm by combining the convolution and pooling process.(2)In terms of classification models: Firstly,is to combine LeNet5 with support vector machine and random forest to construct new classification methods Le＿SVM and Le＿RF for classification experiments.Then,to use convolutional neural networks ResNet and VGG for classification experiments.Finally,the relevant feature extraction algorithm in Chapter 3 of this thesis combined with the classical classification model and the four classification models in Chapter 4 of this thesis are used to classify five hash functions for two,three and four classification experiments.The performance of each feature extraction algorithm and each classification model is compared and analyzed according to the experimental results,and the classification method with the highest overall evaluation is obtained.The experimental results show that among the classification experiments conducted for the five hash functions,the proposed FCN hash feature extraction algorithm in this thesis has more than 75% classification accuracy in all experiments,while the proposed FLLU feature extraction algorithm does not perform as well as the FCN algorithm,but its classification accuracy is nearly 20% higher than the original randomness feature in the binary classification problem and nearly 30% higher in the multi-classification problem 30%.In terms of improved classification methods based on convolutional neural networks,both Le＿SVM and VGG models also perform well,with Le＿SVM’s binary classification accuracy around 90% and triple classification accuracy over 80%.The VGG model,on the other hand,has the best performance among all classification methods with an average accuracy of over 88% for both binary and multiclassification problems.The above experimental data prove that the feature extraction algorithm and the classification method based on convolutional neural network proposed in this thesis are both feasible and effective.