DM_CorrMatch: a semi-supervised semantic segmentation framework for rapeseed flower coverage estimation using UAV imagery

Table 1 Comparison of the accuracy of different quantities of unlabeled data

Dataset (L/U)	Precision(actual/generated)	Recall(actual/generated)	IoU(actual/generated)	F1 score(actual/generated)	Accuracy(actual/generated)
30/300	0.921/-	0.923/-	0.875/-	0.922/-	0.922/-
30/600	0.939/0.936	0.936/0.932	0.884/0.882	0.937/0.934	0.937/0.934
30/900	0.945/0.942	0.948/0.940	0.893/0.886	0.946/0.941	0.945/0.940
30/1200	0.947/0.940	0.948/0.943	0.891/0.888	0.947/0.941	0.947/0.941
30/1500	0.948/0.938	0.951/0.942	0.893/0.885	0.949/0.940	0.947/0.940

Bold values indicate the highest performance achieved when using 30 labeled images and 900 unlabeled images, suggesting that this configuration offers the optimal balance between segmentation accuracy and the image generation efficiency of the diffusion model

ISSN: 1746-4811