Cross- regional Crop Identification Using the Hypothesis Testing Distribution Method

To improve the accuracy of crop classification across temporal and spatial domains. Sentinel-2 satellite images are employed for crop classification training and prediction in select farming areas of Heilongjiang Province by calculating vegetation indices and constructing sequential input feature datasets. The Hunts filtering method was used to mitigate the influence of cloud cover, which increased the stability and completeness of the input feature data across different years. To address the issue of shifts in the input feature values during cross-scale classification, this study proposes the Hypothesis Testing Distribution Method (HTDM). This method balances the distribution of input feature values in the test set even without known crop distribution, thereby enhancing the accuracy of the classification test set. This study utilizes 2019 data on crop planting types from Yushan and Longzhen farms in Heilongjiang Province for model training and data from 10 farms in the province from 2019 to 2022 for model testing. Results indicate that HTDM significantly improves prediction accuracy in cases of substantial image quality variance. After applying HTDM, the recognition accuracy of crop types for the Bawuba Farm in the years 2020 and 2021 reached 95.5% and 96.0%, an increase of 18.2% and 25% compared to before processing, respectively. In 2022, the recognition accuracy for crop types at all farms processed by HTDM was above 87%, showcasing the strong robustness of the HTDM. An analysis of input features using SHAP values revealed that the most impactful features for rice, corn, soybean, and wheat were LSWI in May (LSWI5), LSWI in May (LSWI5), RNDVI in August (RNDVI8), and IRECI in August (IRECI8) respectively.