Abstract: Aiming at the problems of insufficient local feature extraction, high rate of missed segmentation and mis-segmentation of insulator infrared images due to serious noise interference and complex edge information, an insulator infrared image segmentation algorithm based on dual-stream attention and multi-scale fusion is proposed. First, a detail-enhanced convolution module is introduced into the encoder, which further enhances the ability to capture detailed features while retaining the global feature extraction capability; second, a dual-stream attention mechanism is proposed and embedded into the bottleneck structure, which efficiently balances the local and global information to significantly reduce the leakage and mis-segmentation rates; furthermore, a multiscale fusion module is designed in the decoder to improve the the transfer and reuse efficiency of feature information, which minimizes the gradient vanishing problem. The experimental results show that the mIoU of the model reaches 90.93%, the mPA is 95.05%, and the F1 is 94.94%, which are improved by 5.92%, 5.27%, and 2.92%, respectively, compared with the baseline model, which fully verifies the effectiveness of the proposed method.