Recently, object detection algorithms based on convolutional neural networks have been widely applied to insulator detection. However, in foggy scene, existing object detection and two-stage defogging-detection algorithms fail to effectively extract discriminative features, leading to a decrease in insulator detection accuracy. Moreover, the existing channel attention is limited by the receptive field, failing to fully utilize contextual information. This limitation adversely affects the learning of channel weights, consequently diminishing the efficacy of detection outcomes. To address these issues, in this paper, we propose a Dual Constraint Parallel Multi-scale Attention Network for Insulator Detection in Foggy Scene (DCPMA-Net). Specifically, the Contrastive Shared Encoding Dual Constraint (CSED) forms an effective dual constraint by designing the defogging network and detection network shared encoding and the contrastive learning framework between positive samples (insulator) and negative samples (fog and background), which can improve the discriminative ability of the model to extract features in foggy scene. Furthermore, we design a Parallel Multi-scale Channel Attention (PMCA) Module, which extracts multiscale feature information at different stages of channel attention by parallelizing convolutional kernels of different sizes, which can make full use of the multiscale and contextual information to more accurately assign channel weights to the features in the detection network. Experimental results on the Fog Insulator Dataset (FID) surpass those of multiple advanced object detection algorithms. The code and model are available at https://github.com/thishlh/DCPMA-Net.