Aiming at the difficulty of accurately diagnosing gearbox faults with a single convolutional topology of a one-dimensional convolutional neural network in noisy environment,a two-dimensional feature map and deep residual shrinkage network (TM-DRSN) fault diagnosis method was proposed.The fault data samples were obtained based on the overlap sampling method according to the collected gearbox vibration signals,which were divided into training set and test set;based on the lateral interpolation method,one-dimensional data samples were constructed into two-dimensional feature maps which were convenient for DRSN input,and a large convolution kernel layer was constructed in the DRSN input layer as the first feature extraction layer;the residual shrinkage module was added to the deep convolutional neural network to replace the feature extraction layer composed of traditional convolution and pooling;multiple residual shrinkage modules were superimposed to obtain the deep residual shrinkage network model;the constructed DRSN was used in the bearing fault diagnosis test under noisy environment.The results show that the fault diagnosis accuracy of the TM-DRSN method is better than other comparison methods.