Using a Camera System for the In-Situ Assessment of Cordon Dieback due to Grapevine Trunk Diseases

Table 2

Results of training GTD detection algorithm when applied to a completely unseen test set (Block 4).

Identifier	Training data	Training hyperparameters	Trunks detected (%)	Cordons detected (%)	Class accuracy (%)	Variation accuracy (±5%)	Variation accuracy (±10%)
Model 1	V2022 training and validation data. No augmentations applied	See Supplementary Table 1	99.42	15.79	6.43	11.68	14.02
Model 2	V2022 training and validation data. No augmentations. 85% left-right flip applied	See Supplementary Table 2	99.42	82.56	24.12	56.09	71.49
Model 3	V2022 training and validation data. Exposure and blur augmentations applied	See Supplementary Table 3	100.00	88.66	24.99	63.35	77.00
Model 4	V2021 and V2022 training and validation data. No augmentations. 85% left-right flip applied	See Supplementary Table 4	97.09	85.17	26.74	60.74	74.96
Model 5	V2021 and V2022 balanced training and validation data. No augmentations	See Supplementary Table 5	98.84	97.1	25.72	64.39	81.52
Model 6	V2021 and V2022 balanced training and validation data. Exposure and blur augmentations applied. 85% left-right flip applied	See Supplementary Table 6	100.00	99.42	26.16	63.56	84.18