Kiwifruit vine extraction based on low altitude UAV remote sensing and deep semantic segmentation

Abstract

Extraction and segmentation of kiwifruit vine range is an important part of precision agriculture in kiwifruit orchard. In this paper, depth semantic segmentation and traditional machine learning are used to segment and extract kiwifruit vines from orthophoto images, and the accuracy and image quality of vine segmentation based on pspnet, SVM and random forest classification in another test set are compared. Experimental results show that although the mean pixel accuracy of deep semantic segmentation is slightly lower than that of traditional machine learning segmentation, the segmentation image quality of deep semantic segmentation is better and the pixels are more continuous. The deep learning method effectively developed UAV remote sensing image data, improved the usability of remote sensing image, and provided necessary plant information for orchard precision agriculture.