Classification of Flame Combustion State in Rotary Kiln Based on Swin-S Transfer Learning

The flame combustion in rotary kilns is a core process for the smooth and safe industrial production of the entire plant, and real-time classification and identification of its status is a highly challenging problem. However, existing processing and recognition methods have high model complexity, poor timeliness, and their recognition accuracy often fails to meet practical requirements. To address this issue, a rotary kiln combustion state detection system based on the Swin-S transfer learning neural network is proposed. This method uses the windowed self-attention mechanism and shifted window self-attention mechanism in the Swin-S transfer learning neural network to perform localized segmentation and shifted association of the rotary kiln flame combustion images, followed by rapid feature extraction and classification. Transfer learning is introduced by loading a pre-trained model to achieve efficiency and precision. Additionally, pixel block erasure is used to enhance the training sample set, improving the generalization ability of the method. Experimental results show that the recognition accuracy of the rotary kiln combustion state detection system based on the Swin-S transfer learning neural network reaches 97.64%, effectively and rapidly identifying the combustion status within the kiln. Compared with existing methods, the performance is improved, and the system demonstrates good robustness and industrial application prospects.