Fig 1.
Workflow of field-scale detection of BLB in Rice based on UAV multispectral imaging and DL frameworks.
Fig 2.
(a) Pathum Thani Rice Research Center, and (b) Thailand Rice Science Institute.
Fig 3.
Example layout of experimental subplots in the paddy field.
(a) Study area A, and (b) Study area B.
Table 1.
Flight parameters for UAV data collection.
Fig 4.
Field surveying photos in our experiment.
(a) Disease Inoculation, (b) DJI P4 Multispectral, (c) Study area A, and (d) Study area B.
Table 2.
BLB disease scoring for ground truth data collection.
Fig 5.
Sample image and corresponding label used for training the proposed models.
Fig 6.
Framework of the proposed U-NET with ResNet-101 backbone.
Table 3.
Training parameters for model training.
Fig 7.
Performance curves metrics over epochs for proposed models.
Table 4.
Performance of U-Net with ResNet-101 backbone using different combinations of input datasets.
Note that multispectral, multispectral with NDVI, and multispectral with NDRE, models referred to as M1, M2, M3, respectively.
Fig 8.
Comparison of visual classification results with test UAV datasets for BLB disease in rice detection.
Table 5.
Comparison of our proposed model with other models in three different combinations of datasets.