Application Of Granular Computing Thinking In Neural Networks

With the advent of the big data era,uncertain data on large scale and with complex structures is becoming more and more common.The description data of the target object is no longer limited to a single data source.It is necessary to integrate the uncertain data of multiple heterogeneous sources.Therefore,how to quickly and cooperatively analyze these data has become a research hotspot in the field of big data.Granular computing is a model that simulates the thinking of human brain and then solves complex problems.It is a new method,which can simulate the cognitive mechanism of human beings to describe and perceive the real world from multiple dimensions,multiple perspectives and multiple levels,so as to solve complex problems.Granular computing has gradually become an important theory for solving uncertain problems.Multi-granularity analysis has become an important feature of human cognitive ability.Neural network is a parallel processing network model that mimics the formation of neurons.It gives the feedback on real-world interactions by mimicking the biological nervous system.So that it has a strong ability to adapt.Error BackPropagation is an outstanding representative of neural network.It is also a successful neural network learning algorithm.In this paper,back propagation on granular computing is proposed to predict the financial trend.It divides the BP neural network into granularity and makes it with granularity concept in trend prediction,which conforms to the thinking mode of human brain.The BP neural network based on granular computing has obtained high performance in the financial trend prediction.It is more accurate than traditional neural network in predicting stock closing price.Video tracking is a basic task in computer vision.It is widely used in intelligent driving,medical treatment,military engineering and other fields.At present,the research direction is to improve the tracking accuracy through the fusion of multi-type features.However,it fails to consider the mechanism of human target tracking,that is,the granularity relationshipamong multiple characteristics.In order to better simulate the visual tracking mechanism of human brain,the the granular computing thinking is introduced.The relationship between features is described and granulated in a granular way from multiple perspectives and layers.Combined with granular computing thinking,based on deep convolutional neural network.The multi-granular correlation filters video tracking algorithm can consider the granularity relationship among the features,comprehensively analyze the tracking results at each granularity and select the optimal.On the two public datasets of OTB-2013 and OTB-2015,our algorithm is compared with several more advanced video tracking algorithms.It has preferable performance in scale variation,fast motion,in-plane rotation and out-of-plane rotation.In the field of image super resolution reconstruction,granular computing thinking is also applied.By uniting the concept of granular,the multi-granularity deep neural network for super-resolution reconstruction method is proposed.The concept of granularity is introduced in super resolution reconstruction algorithm.Divide the whole image into different granular according to granular size,then carry out the granular size integration after the computation under the granular layer.The proposed algorithm achieves preferable experimental results on the peak signal-to-noise ratio(PSNR).Granular computing combining with neural network provides a new model to solve the problem.In this model,the data,target objects,solving process and even solving results are divided into granularity.Then new operations are performed on the granulated results.Granular computing thinking is a kind of solving thought that conforms to the thinking mode of human brain.It can model and automate the thinking mode of human brain.The research on financial trend prediction,video tracking and image super-resolution reconstruction has verified the correctness of the proposed method.