Image Steganalysis Method Based On Improved Smote And Focal Loss Algorithm
Haotian Zhao, Ke Niu, Zhiqiang Ning, Xiaozhong Pan · Advances in computer science research · 2023
Aiming at the performance degradation of steganalysis model caused by unbalanced sample data set training, a model design method based on improved SMOTE algorithm and Focal loss algorithm is proposed.The improved SMOTE algorithm is used to synthesize new samples to balance the data set.At the same time, the Focal loss algorithm is introduced to pay more attention to the difficult samples and optimize the training process of the model.In the simulation test of the model on BOSSbase1.01data set, under the training of the unbalanced sample set, the detection rate is significantly higher than the similar Zhu-Net method, and the average detection rate is increased by 0.9%, up to 1.9%.This proves the effectiveness of this method and improves the accuracy of model detection.