Yield-SpikeSegNet: An Extension of SpikeSegNet Deep-Learning Approach for the Yield Estimation in the Wheat Using Visual Images

Figures & data

Figure 1. (a) Plant image taken using LemnaTec facility covers not only the plant parts but also the other parts of the chamber, (b) cropped image to get the region of interest (plant parts), (c) ground truth segmented mask images corresponding to the cropped image.

Figure 2. Architecture of Yield-SpikeSegNet consisting of spike detection and yield estimation module.

Figure 3. Training losses of (a) LoPatchExNet, and (b) GlobalPurifyNet.

Table 1. Segmentation performance.

Mean Segmentation error (Err I)	Type II error(Err II)	Intersection over Union (IoU)	Accuracy	Precision	Recall	F1Score
0:0017	0.0562	0.0562	0.9987	0.9982	0.9979	0.9992

Precision=TP/(TP+FP); Recall=TP/(TP+FN); Accuracy=(TP+TN)/(TP+TN+FP+FN); TP: True Positive; TN: True Negative; FP: False Positive; FN: False Negative.

Figure 4. Fitted ANN model architecture.

Table 2. Performance of ANN and SVR in training and testing dataset.

	ANN		SVR
Performance parameter	Training	Testing	Training	Testing
MAE	4.048552	3.851167	3.87128	3.87211
REMS	5.382943	5.223983	5.227	5.202683

Table 3. Summary statistics of the hyper-parameters used in SVR model development.

Cost	Kernel width	Insensitivity	SVR-Kernel	SVR-type	Number of Support ector
1	0.5	0.1	Radial	Eps-regression	609