| Ceramic shuttle kiln is a kind of intermittent kiln applied to the production of ceramics,the temperature detection method of its sintering zone directly affects the quality of ceramic product.The temperature change in the kiln is closely related to the flame burning condition,so the sintering condition of the shuttle kiln can be judged by observing the combustion state of the flame.So far,sintering condition of ceramic shuttle kiln is still identified by thermocouple detection and manual fire watching method which is with low thermal detection accuracy and high labour intensity,and make the automation level of the kiln production low and affect the kiln production efficiency.Therefore,This paper proposes to use flame image recognition to replace traditional thermocouple detection to improve detection accuracy.For the problem of flame image recognition in sintering zone of shuttle kiln,a flame image recognition method based on improved convolutional neural network is proposed in this paper.This method can not only improve the accuracy of flame image recognition of ceramic shuttle kiln,but also effectively improve the detection automation and intelligence level of sintering zone working condition of ceramic shuttle kiln,which is with important theoretical significance and application value.The main work and innovation points of this paper are as follows:（1）A filtering method of improved median filtering combined with bilateral filtering is given,in which the improved median filtering method is first used for smooth denoising preprocessing,and then the improved bilateral filtering method is used for denoising,so that the advantages of both filtering denoising methods are fully utilized without destroying the image edge information,and at the same time effective denoising is achieved.（2）A FCM segmentation algorithm based on energy noise detection is proposed for ceramic shuttle kiln flame image.By introducing the energy curve and clustering the gray level of the image pixels directly,the segmentation quality is improved and the image details are not lost,so the segmentation effect is relatively good.（3）An improved and optimized convolutional neural network method for flame image recognition in the sintering zone of ceramic shuttle kilns is proposed.Based on the Inception-Resnet-V2 model,this method first optimizes and improves the network structure,and then embeds SE module to enable the convolutional neural network to automatically obtain important information of each feature channel,to extract useful features and to suppress those that are not useful for the current task,so as to improve the network recognition rate.The experimental results show that,compared with other methods,the method proposed in this paper has higher accuracy and is effective and feasible. |
