This study presents a deep learning approach for predicting crop yields of maize, rice, and soybeans in Northeast China using multi-source data. The CNN-LSTM-Attention model was evaluated against traditional models such as Random Forest (RF), XGBoost, and CNN. The model integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and an attention mechanism to effectively process complex datasets and capture spatial and temporal variations in crop yields. The study also explores the potential of the Kernel Normalized Difference Vegetation Index (kNDVI) for predicting yields of multiple crops, highlighting its effectiveness. The results show that the CNN-LSTM-Attention model outperforms traditional models in terms of prediction accuracy, with R-squared values of 0.80 for maize, 0.76 for rice, and 0.78 for soybeans. The model's performance is further enhanced by its ability to integrate temporal factors and focus on features most influential for yield prediction. The study also reveals spatial patterns in yield predictions, with the CNN-LSTM-Attention model showing a more refined and accurate distribution of yields compared to other models. The findings demonstrate that advanced deep learning models significantly enhance yield prediction accuracy and provide valuable insights for precision agriculture and food security.This study presents a deep learning approach for predicting crop yields of maize, rice, and soybeans in Northeast China using multi-source data. The CNN-LSTM-Attention model was evaluated against traditional models such as Random Forest (RF), XGBoost, and CNN. The model integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and an attention mechanism to effectively process complex datasets and capture spatial and temporal variations in crop yields. The study also explores the potential of the Kernel Normalized Difference Vegetation Index (kNDVI) for predicting yields of multiple crops, highlighting its effectiveness. The results show that the CNN-LSTM-Attention model outperforms traditional models in terms of prediction accuracy, with R-squared values of 0.80 for maize, 0.76 for rice, and 0.78 for soybeans. The model's performance is further enhanced by its ability to integrate temporal factors and focus on features most influential for yield prediction. The study also reveals spatial patterns in yield predictions, with the CNN-LSTM-Attention model showing a more refined and accurate distribution of yields compared to other models. The findings demonstrate that advanced deep learning models significantly enhance yield prediction accuracy and provide valuable insights for precision agriculture and food security.