U-Net Assisted Farmland Detection with Convolutional Block Attention Module

In the rapidly evolving field of artificial intelligence, computer vision has emerged as a particularly active area. Among the various challenges in computer vision, image segmentation is notably influenced by model performance, especially when applied to complex scenarios involving remote sensing data. Traditional methods often struggle with the inherent high spatial complexity and nonlinear relationships in remote sensing data, frequently leading to suboptimal segmentation accuracy. This research presents an enhanced deep learning model that integrates the Convolutional Block Attention Module (CBAM) with the U-Net architecture. By sequentially applying attention maps in the channel and spatial dimensions, the model improves focus and feature refinement, thereby enhancing the segmentation capabilities of the U-Net framework while maintaining end-to-end trainability. This integration, by emphasising important features, not only enhances feature extraction capabilities but also minimises the impact of irrelevant information, significantly improving the segmentation process. Empirical evaluation on a remote sensing image dataset shows an average Intersection over Union (mIoU) of 86.30%, which is a notable improvement over traditional semantic segmentation techniques and other recent models that incorporate different attention mechanisms.