Abstract: Defect texts accumulated in the operation and maintenance of power equipment may guide the condition evaluation work and overhaul work. However, the complex structure and strong background noise of the defect records lead to the difficulty of information mining intelligently. To address this problem, this paper proposes a dual-channel semantic enhancement network model based on defect text mining. Firstly, the content of the defective text is analyzed, and the defect text is pre-processed by the methods of natural language processing. And the Glove word vector embedding model is used to map the defect text to the numerical space to express the semantics. Then the enhanced text of the defect text is constructed based on word moving distance, and the defect text and enhanced text features are extracted by a bidirectional long-short term memory neural network with an attention mechanism. The key information is enhanced by feature fusion at the end of the network to improve the model effect of classification. The examples show that the classification Macro-F1 metrics of the proposed dual-channel semantic enhancement network are at least 6.2% and 5.2% higher than those of traditional machine learning methods and single-channel deep learning methods, and the proposed method provides a new idea for feature enhancement of multi-source operational data such as images and text.