Automatic detection of agricultural field boundaries from airborne LiDAR data

Object detection from satellite imagery, aerial photographs, and LiDAR data has attracted increasing attention in recent years. In particular, given the growing global concerns regarding food security and sustainable agricultural production, the automatic detection and delineation of agricultural land boundaries have become critically important. LiDAR technology enables the automated acquisition of high-accuracy three-dimensional (3D) point coordinates, which represent both terrain morphology and above-ground objects. For this reason, airborne LiDAR data were adopted in this study. Initially, 10 datasets were generated from airborne LiDAR point clouds. Intensity images derived from these point clouds were subsequently used to extract agricultural parcel boundaries. The intensity images were first processed using specific filtering techniques to reduce noise. Then, K-Means, Fuzzy C-Means, and Meanshift clustering algorithms, together with Canny, Sobel, Roberts, and Prewitt edge detection operators were applied and the most suitable parameter values for each method were systematically determined. The performances of the clustering and edge detection approaches were comparatively evaluated. Accuracy assessments were conducted using manually digitized reference datasets. The effects and performances of several accuracy metrics, including the F1-Score, Peak Signal-to-Noise Ratio (PSNR), Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), Structural Similarity Index Measure (SSIM), and Correlation Coefficient (CC), were analyzed. The experimental results indicate that an average F1-Score of 86% was achieved, demonstrating the effectiveness of LiDAR-derived intensity images for the automatic detection of agricultural field boundaries