An Efficient Transfer-Learning based Approach for Crop Yield Prediction using Remote Sensing Data

One of the most important ways in which humans engage with their natural surroundings is through crop production, which involves the cyclical translocation of carbon across the ecosystem. The computational efficiency remains a big barrier to the algorithm’s broad use, despite its benefits in mechanism and yield estimation robustness. This paper proposes a new hybrid approach to winter wheat yield estimation that combines the Crop-Biomass-Algorithm of Wheat (CBA-Wheat) with the Simple-Algorithm-for-Yield (SAFY) model and transfer-learning(TL). The approach enables calculation efficiency and acceptable accuracy. Using what they’ve learned from the SAFY model, the transfer learning algorithms can forecast wheat harvests. The main results demonstrated that: (1) With R2 of 0.83 and RMSE of 1.91 tha-1, respectively, the comparison of measured and anticipated AGB utilizing CBA-Wheat shows a virtuous correlation; (2) Compared to the other two yield estimation tests, TL with the ample recreation dataset expressively uses significantly less time. With a total of around 16,000 pixels to model, we needed to determine how well the data assimilation technique and TL performed in the absence of full simulation datasets. If the TL model can make production estimation more efficient, it will be a huge boon.