Abstract: Aiming at the insufficiency of the generalization ability of open-loop models and the drawbacks of deep neural network structure for the existing detection methods of the insulator self-exploding state, drawn on the experience of migration learning and closed-loop control, this paper explores an intelligent cognition method of self-exploding state of glass insulator based on deep migration learning, to imitate human cognition model. Firstly, for the pretreated glass insulator images, the interlaced group convolution strategy was employed to reconstruct the convolution layer of Goog Le Net network, which reduced the complexity of network convolution. Secondly, based on the adaptive convolution module group, the data structure of dynamic feature space of the insulator images was built with certain mapping relationship from global to local, and then the discriminative measure index was used to evaluate the difference cognition information of the feature space to enhance the interpretability of the compact feature space. Thirdly, the compact fully connected feature vector was sent to stochastic configuration networks（SCNs） with universal approximation ability to establish the insulator image classification criterion with strong generalization ability. Finally, imitating human thinking mode, based on the generalized error and entropy theory, the evaluation index of the objective optimization function with entropy form for the uncertain cognition results of glass insulator images was built, to real-time evaluate the cognition results of insulator self-exploding state. The dynamic migration learning mechanism is constructed to realize the self-optimization adjustment and reconstruction of the multi-level differentiated feature space of glass insulator self-explosion state and its classification criteria. The experimental results show the feasibility and effectiveness of the proposed method.