基于轻量化网络的陶瓷辊道窑火焰图像识别方法

Flame Image Recognition Method of Ceramic Roller Kiln Based on Lightweight Convolutional Neural Network

针对热电偶在陶瓷辊道窑高温环境下容易腐蚀老化,导致其温度检测精度下降的问题,提出采用计算机视觉火焰检测代替热电偶的陶瓷辊道窑温度智能检测方法。该方法是一种基于新型轻量化卷积神经网络,通过减少网络深度,以及采用大卷积核的设计,避免模型的过拟合现象,减少了模型的复杂度和推算时间,使模型实现了检测准确性的提升。实验结果表明:该新型轻量化卷积神经网络相比原始MobelinetV2模型,识别特征平均准确率提升了4.58%、平均相对误差减少了0.0918%、运算时间减少了41.89%。显然,该新型网络模型使运算速度和分类性能大大提高。

The thermocouple is easy to corrode and age in the high temperature environment of ceramic Roller Kiln,which leads to the decrease of its temperature detection accuracy.This paper proposes an intelligent temperature detection method of ceramic Roller Kiln based on computer vision flame detection instead of thermocouple.The method can improve the detection accuracy based on a new lightweight convolutional neural network model,which avoids over-fitting of models by reducing network depth and using a large convolutional kernel design so as to reduce model complexity and estimation time.Experimental results show that compared with the original MobelinetV2 model,the novel lightweight convolutional neural network proposed in this paper improves the average accuracy of feature recognition by 4.58%,reduces the average relative error by 0.0918%and the operation time by 41.89%.Obviously,the computational speed and classification performance of the new network model are greatly improved.