Agricultural Field Segmentation and Corn Yield Prediction Based on DeeplabV3+ and XGBRegressor

Agricultural field segmentation and crop yield prediction can enhance the efficiency of agricultural production and decision-making. However, current methods often rely on labor-intensive manual sampling, resulting in low efficiency and significant workload. This paper introduces the semantic segmentation model Deeplab V3+to accurately segment remote sensing images containing corn, tobacco, barley, and background. The quantitative relationships between corn crop characteristics and yield are explored by using a variety of machine learning models, including MLPRegressor, HuberRegressor, Linear Regression, Decision TreeRegression, and XGBRegressor. The results showed that DeeplabV3+, with Resnet as its backbone network, could accurately segment agricultural plots, with a loss reduction to 0.2. Notably, XGBRegressor outperformed other methods in the precise prediction of corn yield, with a prediction error of 0.32. This research contributes to the effective monitoring of cultivation plots and crops, offering a new direction for crop yield prediction.