This research focus on accurate yield predictions which are useful for implementing precision farming technologies and making better decisions in crop management. Most of the research has developed a prediction based on crop yield prediction, but failed to consider in increasing computation cost, poor real time use, and overfitting issue. Hence, this proposed model aims to reduce the computation cost, real-time use and reduced the overfitting issue by deep learning based on crop yield prediction. Initially collected the dataset for CropNet data’s are including NDVI, EVI, soil moisture, and vegetation indices, which are pre-processed through atmospheric correction, cloud masking. Following that pre-processing stage for different techniques in image use Anisotropic Mean filter (AnMF) and data use in normalization. Here the mean filter is integrated with anisotropic filter to reduce the noise in the spectral image. The crop yield prediction in different aspects for spatial and temporal attention. The spatial attention in Convolutional Dense EfficientNet with Axial attention (CD-EA) for high dimension spatial features, and temporal attention in Bidirectional coordinate attention based long short layer enclosed Absolute Transformer Encoder model for temporal aware feature representation. The features are integrated via weighted attention based multilayer perceptron (WA-MLP) that provide a prediction results. Also learns vegetation trends and environmental interactions over time, enabling accurate crop yield prediction and effective support for precision agriculture applications. However, the result shows that the proposed framework improved in MSE analysis by 3.895, R2 analysis by 0.9805, MAE analysis by 1.85, RMSE analysis by 1.85 for crop yield prediction are respectively.