Two-phase flow pattern identification method based on multi-scale convolutional network

Abstract

The invention belongs to the technical field of image processing and deep learning, and relates to efficient image classification processing, in particular to a two-phase flow pattern identification method based on a multi-scale convolutional network. The method characterized by at least comprising ten steps. According to the method, an RBF neural network is used for image reconstruction, an image data set of the convolutional neural network is constructed, the data set is divided into a training set, a verification set and a test set according to the proportion of 4: 1: 1, and the test set can be used for network performance testing after multi-scale convolution class network training is completed. Accurate identification of a core type flow pattern and a ring type flow pattern is realized.

Description

technical field [0001] The invention belongs to the technical field of image processing and deep learning, and relates to efficient image classification processing, in particular to a two-phase flow pattern recognition method based on a multi-scale convolutional network. Background technique [0002] The two-phase flow phenomenon widely exists in industrial production processes. As a complex fluid flow phenomenon, it may induce safety problems and even affect the stable and reliable operation of the overall system or equipment. Therefore, obtaining its physical properties is the core that industry and science and technology have been paying attention to. In the study of physical properties of two-phase flow, the study of two-phase flow flow pattern has always been a key point in the industrial production process. With the development of neural networks, traditional BP, wavelet and RBF neural networks have been applied to two-phase flow image reconstruction. Due to the limit...

AI Technical Summary

Problems solved by technology

However, if the network model in this invention is applied to the flow pattern recognition of two-phase flow, there will be the following defects: (1) The image feature extraction is insufficient, and the recognition accuracy is low

But if the network in this invention is applied to two-phase flow pattern recognition, there are following defects: (1) network generalization performance is poor

The network in this invention uses the VGG model, and the network has a large demand for training sample data (at least 10,000), but the image data set for two-phase flow flow pattern recognition is within 1,000, resulting in weak generalization ability of the network. The recognition effect is poor; (2) the VGG model has a large number of layers, reaching 19 layers, resulting in a long training time, so it cannot meet the needs of two-phase flow flow pattern recognition in real time. A fully connected layer is used, which requires a fixed input size, resulting in The network is sensitive to the size of the image and can also overfit

Images