Click on “Download PDF” for the PDF version or on the title for the HTML version.

If you are not an ASABE member or if your employer has not arranged for access to the full-text, Click here for options.

Highlights

A semantic segmentation model is proposed for paddy field images that uses a lightweight backbone network and a boundary enhancement module to enable end-to-end training with annotated images.

Abstract. The precise positioning data for farmland boundaries serves as crucial support for creating high-precision farmland maps, guiding intelligent agricultural machinery in autonomous field operations, and accurately measuring the coverage areas. This study proposes a method for the automatic extraction of farmland boundaries based on low-altitude remote sensing images with Unmanned Aerial Vehicles (UAVs). To effectively segment the boundaries of paddy fields in UAV low-altitude remote sensing images, a semantic segmentation model is constructed, which is based on an improved version of the Mask R-CNN model. The network architecture incorporates MobileNet V2 and Feature Pyramid Networks (FPN) as the backbone network, while the PointRend boundary enhancement module is introduced. Experimental results demonstrate the effectiveness of the proposed model. The experiments were trained and tested on paddy fields in the dataset; the results showed that the mean values of MPA, MIoU, the Average Inference Time per Single Image, and the Quantity of Model Parameters reached 0.9308, 0.8996, 2.6 s, and 6.12 M. When compared to the original model, the improved model brought about an increase in the measures of performance of 0.015 (MPA) and 0.019 (MIoU), while the performance measures decreased by 0.847 (the Average Inference Time per Single Image) and 0.202 (the Quantity of Model Parameters). To address the issue of incomplete paddy field boundary segmentation, a series of morphological operations, including erosion operations and dilation operations, is iteratively applied. The results indicate that the mean pixel accuracy (MPA) of the morphologically processed farmland image is 93.14%, and the mean intersection over union (MIoU) is 89.98%. This iterative process aids in achieving field ridge closure, thereby facilitating the acquisition of boundary information for narrow field ridges and incomplete ridges caused by agricultural machinery.