The text classification task is the most basic and essential in natural language processing. It can find the rules about documents and distinguish the categories of different papers. Nowadays, graph convolutional neural network is widely used in text classification task because of its scalability and high efficiency. However, the GCN model has the best effect at layer 2. As the number of network layers increases, problems such as over smoothing will occur. This paper proposes a semi-supervised text classification model based on a graph attention neural network, which makes full use of the characteristics of information transmission along with the graph by adding residual connections to deepen the number of network layers. Simultaneously, the attention mechanism can be used to give different weights according to the importance of neighboring nodes to target nodes. A dual-layer attention mechanism at the topic level and node level is proposed to learn the importance of different neighboring nodes and extra neighboring node information to target nodes. Experimental results show that this model is superior to other methods on four benchmark datasets.

Text classification is one of the most fundamental tasks in text data analysis. Generally, textual data are extensive in scale and complex in the inherent relationship, leading to low accuracy in traditional classification models. This paper proposes a multi-label classification model based on multi-layer neural network architecture. We first construct a dual-attention mechanism graph neural network (named DGAT for short) to fuse the typological and informational features of the target node and the connected nodes. Secondly, we also build a multi-layer network architecture with multiple DGATs to expand the range of neighborhood nodes participating in the feature fusion to meet the needs of classifying different datasets. To ensure the learning ability of the model, we also use the residual network to solve the problem of error rise and gradient descent caused by multi-layer network architecture. Finally, we conducted a large number of experiments on five benchmark datasets. The results show that the accuracy of the proposed model is significantly better than that of the traditional models, and that there is a noticeable improvement when compared with other deep learning methods.

Effective identification and detection of water-wall tube defects in power plant boiler units are exceptionally signif-icant in reducing the non-stop time of boiler units and avoiding safety accidents. The existing boiler water-wall tube testing technology has some problems, such as dangerous operation, long testing period, high false detection rate, and high testing cost. In this paper, the dataset of boiler water-wall tube defect classification, recognition, and detection is constructed. Then the transfer learning method is adopted to train and compare the classification and recognition model with the target detection model. The optimal model of image defect classification, recognition, and detection are obtained, respectively. Finally, the engineering deployment is carried out. At the same time, we also upload the image classification and recognition module and target detection module to the Web client to display the classification and recognition and detection results of the algorithm server-side in real-time on the Web client. Through the comparative analysis of the experiment, the ResNet101 network is more effective in extracting and learning the defect features of boiler water-wall tubes and better classification and recognition effect on boiler water-wall tubes. The YOLOV5 network has fewer parameters in extracting and learning the target features, which is better than other models.