As a raw material in the iron sintering process, accurate detection of the moisture content of iron ore powder is the prerequisite for controlling the moisture level in sintering mixtures, thereby ensuring optimal air permeability during the sintering process and the quality of sintered ore. Existing detection methods, such as microwave, infrared, and neutron diffraction, exhibit certain limitations, which hinder their effectiveness in field applications and constitute a significant bottleneck in controlling the moisture content of the sintering mixture. Preliminary experiments indicate a strong correlation between the moisture content of iron ore powder and its image texture, which can be characterised by features such as standard deviation, entropy, kurtosis, and peak height. Based on this observation, a moisture content recognition model based on ore powder image texture has been proposed. The model incorporates the Convolutional Block Attention Module into the residual neural network (ResNet) neural network to enhance its ability to extract global features. Additionally, the Modified Local Binary Pattern algorithm is applied to extract local texture feature from the iron ore powder, which are then integrated with the ResNet backbone network to improve the model's focus on local texture. This makes up for the defect of the traditional ResNet model, which tends to prioritise global features, focuses on the extraction of texture features, and realises the accurate identification of iron ore powder moisture. The average precision of the model is as high as 96.31%, the average recall rate is 96.34%, and the average F1-score is 96.53%. Its performance outperforms other mainstream models, including ResNet18, and ResNet50, demonstrating significant potential for practical applications.
Yu et al. (Mon,) studied this question.