Natural gas pipeline leak monitoring is entering the age of big data. Aiming at the problems of traditional methods, such as redundant data, subjective feature extraction and identification, an intelligent pipeline leak aperture identification method is proposed combined compressed sensing (CS) and deep learning theory, which can achieve compressed sampling, adaptive feature extraction and recognition. The random Gaussian matrix is used to acquire the compressed acquisition data, and the aperture information contained in measured samples in CS domain is analyzed by deep learning. The sparse filtering is applied to realize the automatic feature selection. Finally, the high precision classification and recognition of the aperture is obtained by softmax regression. Experimental results show that this method realizes the compression of the monitoring data, and the identification performance for data of compressed sensing domain is better than traditional methods.