부산대학교 도서관

To address the problem that traditional blast furnace air outlet status identification relies heavily on manual work, which is time-consuming and inefficient. In this paper, an artificial intelligence approach is applied to blast furnace air outlet condition monitoring, and a Change-ResNet wind-port condition recognition model based on the improved ResNet18 network is proposed. The proposed model is an integrated model of residual mechanism, which improves the recognition accuracy by adding channel attention mechanism SE blocks to the residual blocks and reduces the number of model parameters by changing the size of the convolutional kernel and the structure of the residual blocks in the model. Introducing SE attention mechanism into the model is to enhance the model's ability to learn and represent key features, thereby improving classification performance. Adjust the residual block structure to reduce the number of parameters and computational complexity in the model, in order to improve the efficiency and practicality of the model. Specifically, firstly, the image dataset is constructed by capturing images by frame from the captured blast furnace air outlet video, and the image dataset is pre-processed; then the residual blocks in the network are used to integrate the high- and low-order feature information, based on which the channel attention mechanism is added to suppress the useless features in the images; finally, the structure of the model and the size of the convolutional kernel are adjusted to reduce the training time of the model. It is proved that the improved method achieves 99.7% accuracy for blast furnace windfall state recognition and the amount of model parameters is reduced by 71.3%.