In order to solve the problem of low accuracy and unstable recognition of fault diagnosis caused by the lack of fault samples of civil aircraft mechanical components and their imbalanced classes, as well as the complexity and diversity of fault signals, a fault diagnosis method based on enhanced meta-learning and channel attention mechanism is proposed. The balanced validation set of small samples is used to guide the updating of the loss weight of the imbalanced training set to improve the distribution of the original imbalanced samples, and the gradient pruning strategy of meta-gradient enhancement is proposed. On the basis of 1D-LeNet, the SE attention mechanism is introduced to adaptively weight the multi-dimensional fault feature channels. The results show that compared with the current mainstream fault diagnosis algorithms such as ProtoNet, DNCNN and GAN-CNN, the proposed method has the best diagnosis effect, and the accuracy is more than 95% when the samples are extremely unbalanced, capable of accurate fault diagnosis.