Aiming at the problem of low recognition rate of small samples by convolution neural network,the concept of confidence is introduced，a new image classification method based on multilayer convolution neural network is proposed, which is called M_CNN,and its application in substation abnormal scene recognition is put forward.According to the network identification of the small sample,set the confidence decision function,picked out the samples which are difficult to identify in the trained single-layer network,re-extracted features and trained the next layer of the network,forming the multilayer convolution neural network structure,so as to achieve the aim of improving recognition performance.The identification results on MNIST database with different sample sizes demonstrate that M_CNN model has some superiority in identifying small samples,at last,the M_CNN model is applied in substation abnormal scene recognition and achieve pretty results.