Figure 3.

Automated root image analysis software. (a) DIRT [40] measures traits based on the ‘shovelomics’ approach [37]. Root systems are washed, and imaged from above in front of a dark background. Root systems are separated from background via thresholding, and RSA traits derived from each segmented object. (b) Root-soil segmentation in X-Ray CT [61]. Root and soil pixels are identified via a Support Vector Machine classifier trained on deep-learned features. Images show the ground-truth, original image, and SVM classifier output. (c) End-to-end deep learning for root tip identification [59^•]. A deep network trained on thousands of instances of root tips and negative samples can be passed over an entire image to obtain likely root tip locations.