Automatic detection of agricultural field boundaries from aerial LiDAR data

Object detection from satellite images, aerial photographs and LiDAR data has become popular in recent years. Especially considering that the food supply and the production problems which has become increasingly important in recent years, automatic detection and classification of agricultural areas is very important. By means of LiDAR technology, which has been used frequently in recent years, 3D point coordinates can be obtained. 3D points contain coordinates terrain and terrain objects. For this reason, LiDAR data was preferred in the study. Within the scope of the study, LiDAR data was obtained freely from the online sources that National Science Foundation, Intergovernmental Committee on Surveying and Mapping and NOAA Office for Coastal Management and also from the Körfez Mapping Planning Company in Turkey as offline data. Initially 10 data sets were generated with 3D point cloud data. Afterwards, using images obtained from 3D point cloud data, parcel boundaries in the agricultural area were determined. In the study, Firstly the intensity images were obtained from the point cloud data were passed through the specific filters. After that K-Means, Fuzzy C-Means and Meanshift clustering algorithms and Canny, Sobel, Roberts, Prewitt edge detection algorithms were applied to the intensity images and the most appropriate parameter values for these algorithms have been determined. Then, the success of clustering and edge detection algorithms was compared. Accuracy analyzes were conducted with reference data sets. The effect and performance of the accuracy criteria that F-Score, Peak Signal to Noise Ratio, Signal to Noise Ratio, Mean Squared Error, Structural Similarity Index Measure and Correlation Coefficient were tested on the results. The results showed that an average success rate of 86 % was achieved in the F-Score value.