In recent years, reservoir lithology identification through deep learning has become a key area of research and a notable challenge in seismic exploration. While traditional deep learning methods focus on analyzing neighboring sampling points in the time domain, they often overlook the valuable frequency information present in seismic data. This limitation hinders the accuracy of reservoir prediction and lithology identification. This paper proposes an AdaBoost-CNN method for reservoir lithology identification. Firstly, this method uses the Continuous Wavelet Transform (CWT) to extract time-frequency (T-F) maps from the seismic data. By doing so, neural networks can effectively utilize the frequency information of the seismic data. Then, to address common deep learning issues like overfitting and class imbalance, the method adjusts training sample weights using AdaBoost to regulate the training process and incorporates transfer learning to reduce training times. The performance and seismic lithology prediction of the CWT-integrated AdaBoost-CNN are compared with STFT-integrated AdaBoost-CNN, STFT-CNN, and CWT-CNN. The field testing results suggest that utilizing AdaBoost-CNN with CWT can improve training efficiency and enhance the generalization capability of models. This leads to a increase in accuracy for seismic lithology prediction and improve lateral continuity in 2-D post-stack seismic profile, which better aligns with geological analysis.