Surface defects have a serious detrimental effect on the quality of steel. To address the problems of low efficiency and poor accuracy in the manual inspection process, intelligent detection technology based on machine learning has been gradually applied to the detection of steel surface defects. An improved YOLOv8 steel surface defect detection model called YOLOv8-MGVS is designed to address these challenges. The MLCA mechanism in the C2f module is applied to increase the feature extraction ability in the backbone network. The lightweight GSConv and VovGscsp cross-stage fusion modules are added to the neck network to reduce the loss of semantic information and achieve effective information fusion. The self-attention mechanism is exploited into the detection network to improve the detection ability of small targets. Defect detection experiments were carried out on the NEU-DET dataset. Compared with YOLOv8n from experimental results, the average accuracy, recall rate, and frames per second of the improved model were improved by 5.2%, 10.5%, and 6.4%, respectively, while the number of parameters and computational costs were reduced by 5.8% and 14.8%, respectively. Furthermore, the defect detection generalization experiments on the GC-10 dataset and SDD DET dataset confirmed that the YOLOv8-MGVS model has higher detection accuracy, better lightweight, and speed.

The surface defects of continuous casting slab have an extremely negative impact on the quality and productivity of steel. Detection of surface defects poses significant challenges in real-time production line, such as excessive reliance on human intervention, low efficiency, and limited measurement accuracy. To tackle these challenges, this paper proposes a lightweight deep learning model called CCSNet, which can achieve efficient feature fusion while reducing computational costs. The new model solves the balance between accuracy and real-time performance of slab surface defect detection. Firstly, real-time photos of the slab surface were obtained using the visual inspection platform installed on the continuous casting production line. From these photos, a dataset containing five distinct kinds of defects was produced. Next, the model composition is introduced, and the feature information required for detection is generated by training the dataset. Finally, the suggested model's validity is demonstrated by model testing and comparison with other models. The CCSNet achieves an average accuracy of 90.1% on the self-made dataset, with the model's weight size being only 6 MB. The detection performance surpasses other traditional models, which can work for real-time defect detection under actual conditions.

Aiming to address the issue of low accuracy in industrial network traffic anomaly detection, we propose an improved DeepFM model for multi-type anomaly detection. The dataset undergoes preprocessing, including encoding and non-string numerical operations. The SMOTE-ENN algorithm is utilized to balance the data through oversampling and undersampling. The improved DeepFM model is employed to extract linear, non-linear, and temporal features from the industrial network traffic data. These features are then fed into an anomaly detector classifier constructed based on Softmax to achieve high-performance detection of traffic attacks. The effectiveness of the model is verified using the UNSW-NB15 dataset, with experimental results demonstrating a detection accuracy of 0.95 for DoS attacks, 0.94 for Fuzzers attacks, and 0.92 for Worms attacks, significantly surpassing other algorithms, which confirms the effective utilization of the proposed model for industrial network traffic anomaly detection.

The sintering flame image has fine-grained local flame state feature information and complex global flame state feature information. However, the traditional convolutional neural network is often more sensitive to local features, and it is difficult to extract the global feature information of the flame state, which restricts the expression ability of sintering flame features, resulting in low accuracy in the classification and recognition of the sintering flame state. In response to such problems, a dual-stream network feature fusion classification method based on CNN-Transformer was proposed, which includes two modules: convolutional neural networks (CNN) flow and Transformer flow. Firstly, the CNN block and the Transformer block were designed in parallel. The CNN stream extracts the local feature information of the RGB image of the sintering flame, and the Transformer stream extracts the global feature information of the GRAY image of the sintering flame. Then, the local feature information and the global feature information of the sintering flame state extracted by the dual-stream network was fused using the cascade interactive feature fusion method. Finally, the softmax classifier was used to achieve the classification of sintering flame states. The experimental results show that the flame classification accuracy can reach 96.20%, which is 6%~8% higher than that of the traditional convolutional neural network.

Cracks at the corners of casting billets have a small morphology and rough surfaces. Corner cracks are generally irregular, with a depth of about 0.2–5 mm and a width of about 0.5–3 mm. It is difficult to detect the depth of cracks and the three-dimensional morphological characteristics. The severity of cracks is hard to evaluate with traditional inspection methods. To effectively extract the topographic features of corner cracks, a multi-scale surface crack feature extraction algorithm, based on weighted adaptive gradient descent, was proposed. Firstly, the point cloud data of the corners of the billet were collected by the three-dimensional visual inspection platform. The point cloud neighborhood density was calculated using the k-nearest neighbor method; then the weighted covariance matrix was used to calculate the normal rate of change. Secondly, the adaptive attenuation rate, based on normal change, was fused with the density weight, which can calculate the Gaussian weight in regard to the neighborhood. Gaussian weights were used to obtain the gradient changes between point clouds to acquire the multi-scale morphological features of the crack. Finally, the interference caused by surface and boundary effects was eliminated by DBSCAN density clustering. The complete three-dimensional morphology characteristics of the crack were obtained. The experimental results reveal that the precision rate, recall rate, and F-value of the improved algorithm are 96.68%, 91.32%, and 93.92%, respectively, which are superior to the results from the RANSAC and other mainstream algorithms. The three-dimensional morphological characteristics of corner cracks can be effectively extracted using the improved algorithm, which provides a basis for judging the severity of the defect.

To address the difficulties in detecting anomalous data in underwater sensor networks and the issue of concept drift in streaming data, a method by constructing an LSTM-CNN Fusion network is proposed. The Long Short-Term Memory neural network and the One-Dimensional Convolutional Neural Network are used as the two base learners of the LSCF network. The Adaptive Sliding Window drift detector is utilized to detect whether data drift occurs. The two base learners are trained independently, and their prediction results are combined through soft voting. The performance of the model is validated using two public datasets, with results showing detection accuracies of 97% and 95% for anomalous data, significantly higher than the average detection accuracy of other models.