Cultivated land serves as the fundamental resource for agricultural production. As an essential component in precision agriculture, the rapid and accurate extraction of cultivated land plays a vital role in crop type identification, crop classification, and yield estimation. This study proposes a novel approach to improve the extraction method of cultivated land plots by leveraging the ResNet50 model as the backbone for feature extraction. Through integrating transfer learning and incorporating attention mechanisms, the fully connected layer of ResNet50 is replaced with the comprehensive Unet architecture. Experimental validation demonstrates that the proposed ResNet optimization model achieves significant enhancements in precision rate, recall rate, and F1-score for cultivated land plots extraction, with respective improvements of 6.25%, 5.63%, and 7.38% compared to the traditional Unet model. Thus, this research holds practical significance and provides valuable insights for the application and promotion of deep learning techniques in cultivated land parcel extraction.
|