Deep learning-based image segmentation for instantaneous flame front extraction

Ruben M. Strässle,Filippo Faldella,Ulrich Doll
DOI: https://doi.org/10.1007/s00348-024-03814-z
IF: 2.797
2024-06-06
Experiments in Fluids
Abstract:This paper delves into the methodology employed in examining lean premixed turbulent flame fronts extracted from Planar Laser Induced Fluorescence (PLIF) images at elevated pressures. In such flow regimes, the PLIF signal suffers from significant collisional quenching, typically resulting in image data with low signal-to-noise ratio (SNR). This poses severe difficulties for conventional flame front extraction algorithms based on intensity gradients and requires intense user intervention to yield acceptable results. In this work, we propose Convolutional Neural Network (CNN)-based Deep Learning (DL) models as an alternative to problem specific conventional methods. The pretrained DL models were fine-tuned, employing data augmentation, on a small annotated dataset including a variety of conditions between SNR 1.6 to 2.6 and subsequently evaluated. All DL models significantly outperformed the best-scoring conventional implementation both quantitatively and visually, while having similar inference times. IoU-scores and Recall values were found to be up to a factor 1.2 and 2.5 higher, respectively, with 1.15 times improved Precision. Small-scale structures were captured much better with fewer erroneous predictions, becoming particularly pronounced for the lower SNR data investigated. Moreover, by applying artificially modeled noise, it was shown that the range of image conditions in terms of SNR that can be reliably processed extends well beyond the images included in the training data, and satisfactory segmentation performances were found for SNR as low as 1.1. The presented DL-based flame front detection algorithm marks a methodology with significantly increased detection performance, while a similar computational effort for inference is achieved and the need for user-based parameter tuning is eliminated. It enables a very accurate extraction of instantaneous flame fronts in large image datasets where supervised processing is infeasible, unlocking unprecedented possibilities for the study of flame dynamics and instability mechanisms at industry-relevant conditions.
engineering, mechanical,mechanics
What problem does this paper attempt to address?
### What problems does this paper attempt to solve? This paper aims to solve the difficult problem of extracting instantaneous flame fronts from Planar Laser Induced Fluorescence (PLIF) images under high - pressure conditions. Specifically: 1. **Low Signal - to - Noise Ratio (SNR) image data**: Under high - pressure conditions, due to the collisional quenching effect, the PLIF signal is significantly weakened, resulting in a very low signal - to - noise ratio of the image data (usually between 1.6 and 2.6). This makes it difficult for traditional flame front extraction algorithms based on intensity gradients to obtain accurate results and requires a great deal of manual intervention. 2. **Limitations of traditional methods**: Traditional methods such as gradient - based edge detection and threshold - based segmentation techniques have shortcomings when dealing with low - SNR images. These methods rely on manually setting parameters (such as thresholds) and are sensitive to noise and cannot well capture small - scale structures. 3. **Improve the accuracy of flame front extraction**: Accurate extraction of instantaneous flame fronts is crucial for studying flame dynamics and instability mechanisms. Especially under industrially relevant conditions, accurate acquisition of local quantities (such as flame curvature) is key. To solve the above problems, this paper proposes a deep learning (DL) model based on a convolutional neural network (CNN) for extracting flame fronts from OH - PLIF images. By using the deep learning model, the authors hope to achieve the following goals: - **Higher detection performance**: Compared with traditional methods, the DL model shows significant advantages both quantitatively and visually. For example, the Intersection over Union (IoU) score and the recall rate (Recall) are increased by about 1.2 times and 2.5 times respectively, while the precision (Precision) is increased by about 1.15 times. - **Reduce manual intervention**: The DL model can automatically process large - scale image data sets without the need for users to frequently adjust parameters, thus unlocking research possibilities that were previously difficult to achieve. - **Expand the scope of application**: By introducing artificial noise, research shows that the DL model can reliably process images under lower SNR conditions (as low as about 1.1), further expanding its application range. In summary, this paper is committed to developing a more robust and reliable method for extracting flame fronts to address the limitations of traditional methods under high - pressure and low - SNR conditions and to provide a new tool for combustion diagnosis research.