Farmland Boundary Line Prediction Method for Autonomous Agricultural Machines
The fast and accurate extraction of farmland boundary is the basis for autonomous and safe operation of self-driving farm machines in the field,and it can also provide basic data for the digital management of the farm.The traditional feature extraction of farmland boundary line images has low accuracy and incomplete boundary line extraction.In this study,we constructed a farmland image annotation dataset and proposed a farmland boundary acquisition method based on UAV remote sensing images,designed an improved semantic segmentation model based on DeeplabV3+,tracked the boundary of the binary image using the boundary tracking function,and obtained the fitted boundary line through the least-squares algorithm after eliminating the outlier points.The results of field experiments show that the IoU of the network is 92.78％and 92.69％for crop-covered and non-crop-covered farmland,respectively,the mPA is 93.96％,and the mean vertical error and the mean angular error of the extracted boundary line were 4％and 0.62°,respectively.This study can provide technical support for the positioning and path planning of automatic driving of agricultural machines.
Automated driving of agricultural machineryBoundary extractionSemantic segmentationLinear fitting
万方数据
维普
