A distributed optical fiber sensing system based on phase sensitive optical time domain reflectometer was used to monitor the filter bag in the dust collector in real time. The positioning of the filter bag in the dust collector is realized by designing the laying mode of the optical fiber in the dust filter bag. The optical fiber vibration signals in six types of damaged bags are collected, and the optical fiber vibration signals in these bags are also collected when they are not damaged. The information entropy and correlation coefficient of the optical fiber vibration signal in the filter bag are calculated by wavelet packet decomposition method and the two parameters are combined to form a two-dimensional characteristic parameter. The characteristic difference between the optical fiber vibration signal in the non-damaged bag and the optical fiber vibration signal in the damaged bag under different two-dimensional characteristic parameters is analyzed. The back propagation neural network is trained with the type 3 filter bag signal feature samples as the training samples, and then the six types of filter bag signal feature samples are identified. The results show that the method has higher recognition stability for the six types of filter bags. And the average recognition rate can reach 96.2%, 88.7%, 98.4%, 98.5%, 98.5%, 98.5%, respectively.