Active Sequential Three-way Decision Model Based On Convolutional Neural Network And Its Application

Face recognition has always been a hot research topic in the field of artificial intelligence.With the development of computers and the Internet,many high-precision face recognition algorithms have been proposed.Most of these algorithms are based on a large number of labeled samples,and the costs of misclassification are defaulted to be identical.However,such assumptions are difficult to meet in most real-world scenarios.In order to solve the problem of cost-sensitive face recognition with fewer labeled images,an active sequential three-way decision model based on convolutional neural network is proposed.This model is based on the sequential three-way decision theory that widely used in multi-granularity classification scenarios.In this paper,sequential three-way decision and active learning strategies are applied for dynamic incremental face recognition problem.First,in order to extract decision information from images and construct different levels of granularity,the convolutional neural network is used to extract features as descriptions of images.Second,considering that the delayed decisions should be more unacceptable with the increase of information,a dynamic boundary region idea is proposed based on the sequential three-decision theory.This idea is that the cost of the delayed decisions should change dynamically with the amount of information.This idea is more in line with the human thinking mode,and can make the boundary region shrink rapidly until it disappears during the sequential process.Finally,considering that in many cases it is expensive or even impossible to obtain enough labeled samples,an active learning method is used to select the samples to be labeled.Inspired by the idea of non-maximum suppression,this paper proposes a non-maximum suppression uncertainty sampling strategy.This model uses active learning strategies to gradually acquire labeled images and builds multi-level decision granularity.Finally,the decision cost and training cost are considered to select the optimal decision step.The effectiveness of the model is verified from multiple perspectives on multiple face datasets.Experimental results show that three-way decision has a lower cost than two-way decision.The active learning method can obtain higher accuracy and lower cost by using fewer samples,and the active sequential three-way decision model using non-maximum suppression uncertainty sampling strategy has the best performance.