Research On Intelligent Temperature Detection Method Of Ceramic Roller Kiln Based On Deep Learning

The ceramic roller kiln is a continuous firing kiln,in which products are preheated,fired,and cooled in sequence through the roller in the kiln to complete the firing process.During the firing process,the temperature detection and control method directly affects the firing quality of ceramic products.Traditional temperature detection methods in ceramic roller kilns mainly rely on thermocouple sensors,with decreased accuracy and complete failure in high temperature and high dust environments,which is not conducive to the long-term stable operation of the roller kiln.Therefore,this paper proposes a method for detecting the temperature of the ceramic roller kiln based on deep learning flame image recognition.In order to improve the classification accuracy of flame image feature,based on the deep learning model,this paper proposes various model improvement methods,and analyzes the impact of these methods on the feature extraction of the model through heat maps,so that it provides technical support for the model design and flame image recognition.The main contributions and innovations of this paper are as follows:1.A flame image preprocessing method for the ceramic roller kiln is proposed.In order to greatly reduce the time consumption of the flame image recognition model in the training and inference process and to improve the recognition accuracy,this paper segments the flame image through the Otus threshold method,and then crops the image through contour detection to remove redundant background data,laying a foundation for effective feature extraction of flame image.2.An improved lightweight network model Mobile Net-m was designed.Taking into account the small amount and single feature of flame image data,the complexity of the Mobile Net V2 model was reduced to obtain the Mobile Net-m model.The experiment results show that this model can reduce 93.86% of the parameter quantity,67.30% of the computational load,and 84.48% of the inference time compared with the Mobile Net V2 model while maintaining classification performance.3.A ceramic roller kiln flame image recognition method based on lightweight network is proposed.According to the test on the convolution kernel size of the lightweight network,it is found that a larger convolution kernel can greatly improve the classification accuracy of the model at the cost of increasing the computational complexity,with a maximum improvement of 6.78%.By analyzing the category activation heat maps,it is found that this method can alleviate the interference of the kiln wall contour features in the ceramic roller kiln flame image,thus improves flame image recognition effect.4.A spatial attention mechanism LSA with a large convolutional kernel is designed.The experiments show that compared with other attention mechanisms,LSA is with a better performance improvement effect on the Mobile Net-m model,and its classification accuracy is improved by 3.66%.By analyzing the influence of attention mechanism on the model,it is found that selecting features from the perspective of spatial attention is more conducive to improving the recognition accuracy of ceramic roller kiln flame images.It provides a new idea for intelligent temperature detection of ceramic roller kiln.