Fig. 4

Overview of main data processing steps and used software packages to process the different types of acquired data. A A trainable Ilastik [21] classification model was trained to robustly cover the diversity of background (yellow labels) due to changing background and media color and diversity of foreground (blue labels) such as different plant species appearance and explant color changes during cultivation. B RGB image processing pipeline was developed in Python [22] with OpenCv [23] and PlantCv [24] for batch processing and including the ilastik classification model headless for segmentation. Upper row: RGB image processing workflow included an automated brightness and contrast adjustment by histogram stretching, down-scaling of image resolution from 4054 px × 3040 px to 1014 px × 760 px. Lower row: The trained classifier predicted binary mask of plant pixels rescaled to the original image resolution and applied to the original image for background removal. Exemplary images from monitoring of A. thaliana (Trial A). C For depth data processing, Python with Open3D [25] was used as an essential component to perform RANSAC [18]-based segmentation. Depth data of in vitro grown A. thaliana seedlings (Trial A). Upper row: Day 0 (Media with 10 day old, small seedlings), Hough Transform circle detection [26] and edge-removed depth image. Lower row: Pseudo 3D visualization of depth data of Day 11, estimated RANSAC plane and plant height surface corrected by estimated RANSAC plane at Day 11. Detailed description in “Methods” section