Recognition of gas–liquid two-phase flow patterns based on improved local binary pattern operator Wenyin Zhang a, * , Frank Y. Shih b , Ningde Jin c , Yinfeng Liu a a School of Informatics, Linyi Normal University, Linyi, Shandong 276005, PR China b College of Computing Sciences, New Jersey Institute of Technology, Newark, NJ 07102, USA c School of Electrical Engineering and Automation, Tianjin University, Tianjin 300072, PR China article info Article history: Received 13 April 2010 Received in revised form 21 April 2010 Accepted 9 June 2010 Available online 15 June 2010 Keywords: Local binary pattern Two-phase flow regime Support vector machine Neural network abstract A new method to pattern recognition of gas–liquid two-phase flow regimes based on improved local bin- ary pattern (LBP) operator is proposed in this paper. Five statistic features are computed using the texture pattern matrix obtained from the improved LBP. The support vector machine and back-propagation neu- ral network are trained to flow pattern recognition of five typical gas–liquid flow regimes. Experimental results demonstrate that the proposed method has achieved better recognition accuracy rates than oth- ers. It can provide reliable reference for other indirect measurement used to analyze flow patterns by its physical objectivity. Ó 2010 Elsevier Ltd. All rights reserved. 1. Introduction Nowadays, much more attention has been paid on the changing patterns and structures of multi-phase flows, which are exten- sively encountered in petroleum exploitation and transport, chemical engineering, nuclear reactors, and thermal systems. In the gas–liquid two-phase flow, the distribution of the two media, namely flow structure, is extremely complex and can vary instan- taneously because of stochastic variation of gas–liquid two-phase flow interface. Therefore, it is crucial to understand the interior structure and flow properties of different flow patterns. At present, image and video processing technologies and high frame-rate pho- tographic techniques are used to analyze the inner structure and flow properties of different flow patterns quantitatively and qual- itatively. The aim is to develop a mathematical model and make a deep understanding of different temporal and spatial flow char- acteristics, as shown by Hsieh et al. (1997), Gopal and Jepson (1997), Shi et al. (2005), Sathyamurthi et al. (2007), Bui et al. (1999), Zhou et al. (2009). As a matter of fact, many gas–liquid two-phase flow analysis and recognition methods based on image or video processing tech- niques have been developed in recent years. The dynamic images of two-phase flow patterns obtained by high frame-rate camera are characterized by direct-vision and physical objectivity, and are able to reflect the information about the complex two-phase flow structure to a certain extent. Hsieh et al. (1997) acquired dy- namic images of gas–liquid two-phase flow in the riser in a natural circulation loop by CCD and gave the statistics about image gray- level distribution probabilities of four flow patterns, namely sin- gle-phase flow, bubbly flow, slug flow, and churn flow, to classify them and analyze their dynamic variations by image processing techniques. Gopal and Jepson (1997) studied the dynamic charac- teristics quantitatively, including local velocity and void distribu- tion of slug flow in gas–liquid two-phase mixtures, by applying image processing techniques and the kinetics model of slug flow on the experimental conditions. Shi et al. (2005) presented image analysis to extract bubble features of gas–liquid two-phase flow images and employed the fuzzy inference algorithm to identify flow patterns. Sathyamurthi et al. (2007) applied high-speed dy- namic photography and chaotic box-counting techniques to calcu- late fractal order of bubble voids in nucleate boiling wall, which consists of an array of individually controlled micro-heaters. Bui et al. (1999) presented a new method for automatic bubble identi- fication of two-phase bubbly/slug flow. Recently, Zhou et al. (2009) used histogram to compute the flow image features and applied the support vector machine (SVM) to identify the flow patterns. Although there are great progresses in studying the two-phase flow properties, most of the research schemes are based on single and standard image of typical flow pattern preprocessing to extract some characteristic parameters of flow patterns. In the two-phase flow process, the temporary flow patterns are various at different positions as a result of the stochastic variability of phase hold- up, flow structure, and the interfacial interaction. Therefore, it is 0301-9322/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijmultiphaseflow.2010.06.002 * Corresponding author. E-mail address: zwy218@hotmail.com (W. Zhang). International Journal of Multiphase Flow 36 (2010) 793–797 Contents lists available at ScienceDirect International Journal of Multiphase Flow journal homepage: www.elsevier.com/locate/ijmulflow