Abstract: To adapt to the operation characteristics of the new distribution network, the distribution network switches require frequent adjustments to their structures. However, it is difficult to timely and accurately obtain the real-time topology of the distribution network, which poses challenges for situational awareness of the network. Traditional topology identification methods based on state estimation are difficult to apply online due to their high computational complexity and the large number of topology categories in large-scale distribution network. To address these challenges, this paper proposes a distribution network topology identification method based on multi-label classification and convolutional neural network (CNN). By exploring the multi-mapping relationship between measured voltage data and switch states, a multi-label classification mechanism is introduced to encode the distribution network topology. The switches are physically mapped to the topology identification model output and a CNN is used to build a multi-label classifier, achieving accurate topology identification. Verification of the proposed method is conducted using a revised IEEE 123-node distribution network, and experimental results show that it has a high topology recognition accuracy. Additionally, the model demonstrates better inference capability for unknown topologies outside the training sample space, making it more suitable for practical topology identification scenarios. The superiority and robustness of the proposed method can be verified.