Intelligent Monitoring Method of Crude Fuel Images Based On Deep Learning

Abstract

The conditions of crude fuel can reflect the reaction degree in a blast furnace, and the real-time monitoring and analysis of the crude fuel can improve production performance and stabilize the conditions of the furnace. The results of crude fuel conditions obtained by traditional manual sampling detection are low in accuracy, and have danger and hysteresis. In order to reduce the workload of personnel and improve detection accuracy and timeliness, this paper proposes an intelligent monitoring method of crude fuel images based on deep learning. According to the method, attention mechanisms are added on the basis of a Mask R-CNN algorithm, so that the detection accuracy is improved, and besides, the problem of overfitting is solved. In order to ensure the detection accuracy under high-speed motion blurred images, a DeblurGAN-v2 algorithm is used to deblur the images; and when a dataset is built, data enhancement is used to increase the number and types of samples, so that the algorithm can adapt to the actual production environment of a factory. Through a crude fuel detection experiment, the effectiveness of the algorithm in the aspect of improving the detection accuracy of clear and blurred images is verified.