Aiming at the fact that rolling bearing vibration signals are susceptible to non-stationary noise,a multi-scale kernel network with improved attention mechanism (IA-MKNet) was proposed to extract more sensitive characteristic signals from noisy vibration signals.An improved multi-scale convolutional attention mechanism (IAM) was proposed to adaptively extract meaningful fault features and automatically suppress noise. Then,for the inherent multi-temporal characteristics of vibration signals,an adaptive multi-scale kernel based on IAM was designed.Residual blocks were used to capture the multi-time-scale fault features of vibration signals.Finally,a combination strategy based on adaptive ensemble learners was proposed to increase the diversity of features by fusing the outputs of multiple IA-MKNets,thereby further improving the diagnostic accuracy and stability.The experimental results show that the method improves the fault diagnosis accuracy of rolling bearings in a noisy environment,and its performance is better than the other five benchmark methods.