Plant Phenomics / 2019 / Article / Fig 2

Research Article

A Weakly Supervised Deep Learning Framework for Sorghum Head Detection and Counting

Figure 2

Automated annotation protocol: the initial input image passed through the network is part of the training dataset. This trains the RetinaNet architecture and generates a model, which then is used to annotate new data. This annotation is then verified by human expert raters and fed back into the network as new training data. Process is repeated until network reaches desired level of accuracy.