Fig. 10
From: Segmentation and counting of wheat spike grains based on deep learning and textural feature
Grad-CAM visualization results highlight the important regions of the training model predicted. We compared the visualization results of CBAM-HRNet and HRNet. The Grad-CAM visualization was calculated on the final convolutional output. Correctly predicted categories are shown in red and incorrectly predicted in blue. CBAM supports the network in correcting the predictions and improving the effect of target segmentation