Automatic Segmentation of Crops in UAV Images

dc.contributor.authorZheng, Yuanyuan
dc.description.abstractRemote sensing imagery has been increasingly utilized in agricultural production due to its convenience and cost-effectiveness. However, traditional methods for crop segmentation require significant time and manual effort. Therefore, this research proposed the use of threshold segmentation and deep learning techniques to achieve automatic crop segmentation in UAV images and evaluated their performance. Specifically, this research utilized image threshold segmentation, a custom UNet network, Deeplabv3+ and segment anything model(SAM) with multiple prompts. The results showed that the Intersection over Union (IoU) for threshold segmentation was 0.58. The IoU for UNet was 0.70, and for DeepLabV3+ it was 0.76. The IoU achieved by SAM with points prompt was 0.89, demonstrating superior crop segmentation performance. However, the masks generated using SAM automatic mask generation and a bounding box with a point prompt couldn’t segment crops effectively.en_US
dc.rightsAttribution 4.0 International*
dc.titleAutomatic Segmentation of Crops in UAV Imagesen_US
dc.typedissertation or thesisen_US


Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
2.94 MB
Adobe Portable Document Format