Given that the manual feature selection process is troublesome and not sufficiently accurate, an integrated learning method based on Gramian Angular Field (GAF) and optimal feature channel adaptive selection is proposed when designing the rolling bearing fault diagnosis model. First, the GAF transformation is performed on the raw vibration signal compressed by using the Piecewise Aggregation Approximation (PAA) technique, and the vibration signals of different states are encoded into different types of GAF images. Then convolutional channel attention residual network (CCARN) is used to learn advanced features from GAF images and the fault diagnosis results are produced. Finally, in order to further improve the stability of the proposed fault diagnosis method, an integrated learning method based on hierarchical scoring strategy is proposed. The reliability of the output results of diagnostic model is further enhanced through the voting decision results of multiple models. Experimental results show that the proposed method has good classification performance on rolling bearing data, and outperforms those state-of-the-arts.