The DL model (DCNN) for phenotyping pest resistance as binary and continuous traits. (IMAGE)
Caption
The DL model (DCNN) for phenotyping pest resistance as binary and continuous traits. A, Performance of six classic convolutional neural networks in CV. The bars correspond to the left Y-axis (accuracy score), while the line is referenced to the right Y-axis (model complexity). B, The number of correct recognitions of different categories by VGG16 in the test set. C, The model architecture of DCNN-PDS and the workflow for deriving continuous traits from images. The model uses pretrained VGG16 for initial feature extraction, followed by four residual blocks (each consisting of two convolutional layers) to enhance deep feature learning and prevent degradation. A global average pooling (GAP) layer is then applied to reduce the feature dimensions, and three fully connected layers are used for feature fusion. D, Examples of the DCNN-PDS prediction results for the different extent of damages by pest, and the PDS values are shown at the bottom of each plot. E, The correlation between the predicted values provided by DCNN-PDS and the manually labeled values in the test set.
Credit
Horticulture Research
Usage Restrictions
Credit must be given to the creator.
License
CC BY