Deep Fusion with Attention Neural Network for Soil Health based Crop Yield Estimation using Remote Sensing Images

Crop yield estimation, vital for agricultural planning, incorporates weather, soil health, and technology. Utilizing remote sensing to analyze soil health enhances agricultural management and resource optimization. Despite challenges like data accuracy and cloud interference, the proposed multi-head cross attention with capsule energy valley network (mhca-cevn) tackles these issues. This research integrates sentinel-1 and sentinel-2 data with field measurements, employing advanced preprocessing and feature extraction methods, such as the guided multi-layer side window box filter and shearlet transform. The hybrid gold rush mantis search optimizer selects key features for a deep visual attention-based fusion method. The resulting mhca-cevn classification model achieves outstanding performance, with accuracy, sensitivity, error rate, f1-score, mean absolute percentage error, and symmetric mean absolute percentage error at 97.59%, 95.21%, 6.65%, 90.21%, 5.01%, and 0.042%, respectively. These metrics highlight the model's efficacy in addressing diverse crop yield challenges, establishing it as a robust solution for remote sensing.