A large number of defect images and textual data have been accumulated during the operation and maintenance of power equipment. These data are crucial for guiding fault diagnosis and maintenance decisions for power equipment. To address issues such as the singular data modalities

low-level feature fusion

and poor data quality in existing power equipment defect classification tasks

this paper proposes an image-text fusion classification method based on an improved attention mechanism and contrastive learning (IAC-ITFusion). Firstly

the method introduces a dual-cycle cross-modal attention (DCCA) mechanism to capture the mapping relationships between image and text data while integrating feature information. Secondly

based on the concept of contrastive learning

an attention-guided loss function is proposed to regulate the learning direction of the DCCA mechanism

ensuring it to focus on the correct feature information to achieve effective fusion of image and text features. Finally

experiments were conducted on the image and text fusion classification tasks for power line and substation equipment defects. The results show that the proposed method can achieve accuracy rates of 98.48% and 98.57%

respectively

thus the method's effectiveness in image and text fusion classification tasks for power equipment defects is verified

which is significant for advancing the intelligent development of power equipment operation and maintenance.