Boundary Detection in Agricultural Fields Using a Residual U-Net++ Architecture

Agricultural land boundary detection with satellite imagery is really important for modern farming, especially in precise mapping, resource management, land changes monitoring, and violations controlling in agricultural areas. In order to accurately detecting field boundaries, this paper presents a new deep learning method that combines Residual U-Net and U-Net++ architectures for precise segmentation of farm fields. By mixing residual connections within a nested U-Net setup, the model improves feature extraction and refines field boundary detection. We used a loss function that helps with class imbalance, boosting segmentation performance. Tested on 1200 high-resolution aerial images from Catalonia, Spain, the model shows better results than existing methods, significantly improving field boundary detection. This method advances field segmentation technology and offers a valuable tool for precision agriculture.