Table 1 Comparison of representative studies on agricultural transformation Evaluation.
Author (Year) | Main Contribution | Method | Limitations |
|---|---|---|---|
Srivastava et al. (2022) | Developed CNN model to analyze temporal links between environmental variables and winter wheat yield | 1D convolution to capture temporal relations of environmental variables | Ignored spatial heterogeneity; no metaheuristic optimization |
Mujeeb & Javaid (2023) | Proposed ISDAE-PSO framework to optimize carbon emission prediction | Feature engineering combined with PSO to optimize DNN parameters | Did not consider regional policy and climate differences; single data source |
Guo et al. (2024a) | Proposed DCNN-LSTM hybrid model for multi-index climate prediction | Combined DCNN for spatial features and LSTM for temporal dependencies | No hyperparameter optimization; limited adaptability in dynamic scenarios |
Rokhva et al. (2025) | Built EfficientNetB7-CBAM model for food classification | Attention mechanism fused with transfer learning | Not adapted for agricultural multimodal data; lacks spatiotemporal modeling |
El-Kenawy et al. (2024) | Developed GWO-optimized framework for potato yield prediction | Grey Wolf Optimization integrated with gradient boosting | Did not integrate remote sensing data; policy dynamic effects not encoded |
This study | Constructed CNN-LSTM-SMA hybrid model for multisource-driven dynamic agricultural transformation evaluation | Fused CNN-LSTM to extract spatiotemporal features, optimized hyperparameters with SMA, designed region-adaptive strategies | Dependent on historical data timeliness; cross-large-scale geographic generalization requires further validation |
