Research on manifold recognition methods based on deep learning image data augmentation

In industrial processes such as petroleum extraction and gathering-transportation,the two-phase oil-water flow is ubiquitous and exhibits complex and diverse flow patterns. Effective flow pattern recognition is crucial for optimizing petroleum-related processes, improving production efficiency, and ensuring operational safety. To address the issues of low recognition accuracy and poor model generalization caused by the limited number of flow pattern samples, this study integrates image data augmentation techniques,such as resizing, flipping, and brightness adjustment,to effectively expand the dynamic flow pattern dataset of the oil-water two-phase flow. A deep learning-based flow pattern recognition model is constructed, and the training process is evaluated using two objective metrics. Loss and accuracy (Acc). Comparative analysis of the test results of the original data set with the augmented data set demonstrates that the latter can be expanded to approximately 12 times the original size, resulting in an improvement of 4.96% in flow pattern recognition accuracy. The generalization capability of the flow pattern classification model has been significantly improved, making it more suitable for real-world industrial processes with complex flow patterns.