In order to realize the intelligence of the production, storage and assembly process of prefabricated building components, an online recognition method based on improved view clustering was designed. By adding label information embedding and label information solidification into the information bottleneck algorithm, the traditional view clustering algorithm was improved into an unsupervised pattern recognition method which could be used for online recognition of prefabricated building components. The method was tested on the real data set of an industrial construction company in Tianjin, and the experimental results show that the model recognition accuracy is above 90%, which is superior to Softmax neural network, support vector machine, Bayesian network and other supervised pattern recognition methods.