Description of Invention
This project explores using machine learning to enhance paddy field management through precise weed mapping. Traditional methods rely on manual labor, which is time-consuming. By employing deep learning, this project aims to improve weed detection and classification, facilitating targeted management. We use semantic segmentation models like PSPNet, UNet, and SegNet to classify crops into broadleaved, sedges, grass, and paddy. Data preprocessing includes background removal and data augmentation. Models are trained in two stages: first on simple, single-instance images, then on complex, multi-class datasets. The UNet with ResNet50 achieved 96.33% accuracy, providing valuable insights for sustainable farming practices.