HairNet: a deep learning model to score leaf hairiness, a key phenotype for cotton fibre yield, value and insect resistance

Table 3 Selection of visual feature extraction network based on % accuracy scores

Dataset\(\rightarrow \)	Y1/Y1			Y2/Y2			Whole dataset
Network \(\downarrow \)	IA	FIA	LA	IA	FIA	LA	IA	FIA	LA
ResNet18	94.55	94.87	98.71	84.00	82.00	90.50	83.19	78.41	89.56
ResNet34	94.09	94.87	97.43	84.37	83.50	90.00	84.85	81.65	91.36
ResNet50	94.43	96.15	98.71	82.43	77.50	88.00	82.60	79.85	89.92
ResNet101	93.04	91.02	98.71	82.75	83.00	88.50	82.49	83.09	88.12
ResNet152	93.27	91.02	97.43	80.81	74.50	86.00	82.60	79.85	87.76

For each network structure tested, % accuracy scores are reported as Image Accuracy (IA), First Image Accuracy (FIA) and Leaf Accuracy (LA). The highest accuracy in each column and corresponding feature extractor is highlighted in bold. All models here employ the Adam optimizer with learning rate (lr) \(=1e^{-4}\)

ISSN: 1746-4811