Recognition algorithms have been widely used in brain computer interface (BCI) for neural paradigms classification. To improve the classification and recognition effect of motor imagery with motor observation (O-MI) in BCI rehabilitation technology, this paper explores the function of convolutional neural network (CNN) combined with synchrosqueezed wavelet transform (SST) and long short-term memory (LSTM) in the recognition and classification of neural activities in the brain motor area. Combining the advantages of SST in signal feature extraction in the pretreatment stage and the ability of LSTM network in time series information modeling, the purpose is to make up for CNN's shortcomings in both aspects. This paper verifies the algorithm on the self-collected O-MI experimental datasets and the public datasets (BCI competition IV datasets 2a). The results show that the composite CNN algorithm incorporating SST and LSTM achieves higher classification accuracy than classic algorithms and the similar new method which is CNN combined with discrete wavelet transform (DWT) and power spectral density (PSD), so it is convenient for practical application in O-MI BCI system.