Abstract: To solve the problem of pattern recognition interference caused by signal non-correspondence during partial discharge(PD) photoelectric joint detection of gas-insulated transmission line(GIL), this paper proposes an image fusion algorithm based on non-subsampled shearlet transform(NSST) and sparse representation(SR). The phase-resolved partial discharge(PRPD) patterns and the ultra-high frequency PRPD patterns are firstly decomposed into high-frequency and low-frequency subband patterns by NSST. Then, the subband patterns are fused based on maximum absolute value and SR. Next, the fused photoelectric PD patterns are obtained through inverse NSST transformation. Finally, features of the fused patterns are extracted and feature dimensions are reduced. Moreover, K-nearest neighbor(KNN), support vector machine(SVM), naive Bayesian(NB), and decision tree(DT) classifiers are applied for pattern recognition and accuracy comparison with other algorithms. The test results show that, regardless of the overall size of the sample and relative quantity of training set to test set, the proposed algorithm can preserve relatively complete information, and pattern recognition accuracy is much higher than that of a single NSST algorithm or SR algorithm. In the case of a small sample(150), the algorithm can achieve 89.2% recognition accuracy, which can even be up to 98.5% when the sample is large enough. While the training set is smaller than the test set, the recognition accuracy is still above 70.0%, and can be up to 88.0%, which provides a reference for improving the PD pattern recognition accuracy of GIL.