Sustainable Crop Yield Forecasting with the Dual-Branch Geometric Progressive Feedback Cosine Convolutional Network Model

Using deep learning algorithms to extract important agricultural traits has made crop production prediction based on environmental, soil, water, and crop parameters an important area of research. But conventional approaches have trouble creating a straight linear or non-linear mapping between yield values and raw data, and the quality of the characteristics that are extracted has a significant impact on how well they work. Deep reinforcement learning builds a strong prediction framework by combining the intelligence of deep learning with reinforcement learning to get beyond these drawbacks. In order to anticipate agricultural yields accurately and sustainably, this work aims to create a Dual-Branch Geometric Progressive Feedback Cosine Convolutional Network with Skill Optimization Algorithm (DB-GPFCNet-SOA). Robust Double-Weighted Guided Filtering (RDWGF) is used for data preprocessing in order to minimize noise and improve important components. To ensure a complete prediction model, the N-Branch retrieves local neighbor-based information while the C-Branch records global geometric properties. Moreover, the accuracy is enhanced by the SOA, which enhances loss parameters. The suggested model has RMSE of 0.25 and MAE of 0.15 with exceptional accuracy of 99.2%, precision of 99.1%, recall of 99%, and F1-score of 99%. These results validate the performance of the model, which renders it a suitable approach for real-time agricultural applications.