Figure 2.

Application of deep learning. (A) Neural network architecture. Tuning the dense layers appeared to improve the learning rates. (B) Influence of the number and types of images on learning rates. The validation accuracy was reduced irrespective of the presence (lower right) or absence (lower left) of a green filter. The total number of images was likely more important for increasing the validation accuracy, at least in this small-scale study (upper). Blue line, training-accuracy curve; Green line, validation-accuracy curve.