Fig 1.
Rice seedlings and weeds images in the paddy field.
Fig 2.
(a) original image and (b) the corresponding GT labels.
Table 1.
Number of pixels with classes and the class weight coefficients.
Fig 3.
Network architecture of SegNet.
Table 2.
Encoder and decoder parameters of SegNet.
Fig 4.
Pooling and upsampling.
Fig 5.
Feature map visualization of the convolutional layer.
Fig 6.
Performance comparison on test images.
(a) original images; (b) ground truth; (c) output by our method; (d) output by FCN; and (e) output by U-Net.
Table 3.
Results of SegNet, FCN and U-Net.
Table 4.
Comparison of the SegNet, FCN and U-Net approaches.