基于卷积神经网络的钢渣砂图像识别及图像变化规律

钢渣安定性检验是实现钢渣安全资源利用的关键，针对钢渣安定性检测方法的效率低且受到取样代表性不足的问题，提出一种基于卷积神经网络的钢渣砂图像分类模型SE-ConvNeXt。该分类模型针对钢渣砂的图像特征，在ConvNeXt网络中添加通道注意力机制SE-Net（squeeze and excitation network）。相比于原ConvNeXt和其他卷积神经网络模型，SE-ConvNeXt的收敛速度更快，训练过程更稳定，准确率更高。实验数据集采集于蒸汽处理前后的钢渣砂图像，钢渣砂的粒径为4.75-2.36mm和2.36-1.18mm。分别使用两个粒径的钢渣砂图像训练网络，并分析钢渣砂图像变化规律。模型预测两个粒径的钢渣砂图像数据集准确率分别为92.5%、94%，且两个粒级的钢渣砂图变化规律相似，随着蒸汽陈化时间的增加，变化程度逐渐变小，随后图像变化程度趋于稳定。分析粉化率的变化规律，钢渣砂粉化率变化规律与钢渣砂图像变化规律具有相关性，蒸汽处理的钢渣砂可通过钢渣砂图像评价体积安定性。

Steel slag sand image recognition and image change rule based on convolutional neural network

Steel slag stability test is the key to using steel slag safety resources. They are aiming at the problems of low efficiency and lack of representativeness of steel slag stability detection methods. A steel slag sand image classification model SE-ConvNeXt based on a convolutional neural network is proposed. The classification model adds channel attention mechanism SE-Net ( squeeze and excitation network ) to the ConvNeXt network for the image features of steel slag sand. Compared with the original ConvNeXt and other convolutional neural network models, SE-ConvNeXt has a faster convergence speed, a more stable training process, and higher accuracy. The experimental data set was collected from the images of steel slag sand before and after steam treatment. The particle sizes of steel slag sand were 4.75-2.36 mm and 2.36-1.18 mm. The steel slag sand images of two particle sizes were used to train the network and analyze the change rule of steel slag sand images. The model predicts that the accuracy of the steel slag sand image data sets of two particle sizes is 92.5 % and 94 %, respectively. The change rules of the steel slag sand images of the two-particle sizes are similar. As the steam aging time increases, the degree of change gradually decreases, and then the degree of image change becomes stable. By analyzing the change law of pulverization rate, the change law of pulverization rate of steel slag sand is related to the change law of steel slag sand image. The volume stability of steam-treated steel slag sand can be evaluated by the steel slag sand image.