Development and Validation of a Deep Learning Based Automated Minirhizotron Image Analysis Pipeline

Abstract

Root systems of crops play a significant role in agroecosystems. The root system is essential for water and nutrient uptake, plant stability, symbiosis with microbes, and a good soil structure. Minirhizotrons have shown to be effective to noninvasively investigate the root system. Root traits, like root length, can therefore be obtained throughout the crop growing season. Analyzing datasets from minirhizotrons using common manual annotation methods, with conventional software tools, is time-consuming and labor-intensive. Therefore, an objective method for high-throughput image analysis that provides data for field root phenotyping is necessary. In this study, we developed a pipeline combining state-of-the-art software tools, using deep neural networks and automated feature extraction. This pipeline consists of two major components and was applied to large root image datasets from minirhizotrons. First, a segmentation by a neural network model, trained with a small image sample, is performed. Training and segmentation are done using “RootPainter.” Then, an automated feature extraction from the segments is carried out by “RhizoVision Explorer.” To validate the results of our automated analysis pipeline, a comparison of root length between manually annotated and automatically processed data was realized with more than 36,500 images. Mainly the results show a high correlation (r=0.9) between manually and automatically determined root lengths. With respect to the processing time, our new pipeline outperforms manual annotation by 98.1-99.6%. Our pipeline, combining state-of-the-art software tools, significantly reduces the processing time for minirhizotron images. Thus, image analysis is no longer the bottle-neck in high-throughput phenotyping approaches.

Uncovering natural variation in root system architecture and growth dynamics using a robotics-assisted phenomics platform

“…Recent progress in machine learning is demonstrating the opportunities within the field for segmenting out root and noise and could further improve accuracy ( Wang et al, 2019 ; Smith et al, 2018 ). Whereas RGB imaging together with software tools like RootPainter has facilitated automated root system analysis in species with thicker roots ( Smith et al, 2020 ; Bauer et al, 2022 ), bioluminescence still has advantages with species with finer roots, like Arabidopsis , since it improves the contrast between the root and soil signal. Using transgenic lines homozygous for the luciferase transgene would lead to stronger signal and therefore better root detection.…”

High Frequency Root Dynamics: Sampling and Interpretation Using Replicated Robotic Minirhizotrons

“…We extracted RLD and RSA in an extremely simple fashion from segmented images, which contained noise but was compared with a similarly coarse above-ground index. More interpretive root phenotyping (Bauer et al ., 2022) can also be paired to this CNN. In theory, many further architectural traits can be extracted with such a workflow.…”

“…Minirhizotron imagery contains much more relevant structural information due to the extremely local scale and indeed these are regularly extracted manually. There is progress in doing this automatically (Seethepalli et al ., 2021; Bauer et al ., 2022) and we expect rapid advancement in future. Our analyses were also not particularly sensitive to roots overlapping or growing close due to overall low density.…”

“…We did not adjust for diagonal connections nor prune branches of the skeleton but otherwise this is similar to the simplest level of analysis using modern software for root imagery such as RhizovisionExplorer (Seethepalli et al ., 2021). This software can be directly paired to the rootpainter CNN and has been successfully use to process imagery from agricultural soils with similar accuracy to manual annotation (Bauer et al ., 2022) and can potentially extract many more features of interest. However, our datasets in natural ecosystems were both challenging to the CNN segmentation and for the human operator to choose appropriate settings for the software.…”

“…Recently, convolutional neural networks (CNNs) showed promising results to identify roots in a variety of settings (Delory et al ., 2016; Rahmanzadeh and Shojaedini, 2016; Vincent et al ., 2016; Huo and Cheng, 2019; Bauer et al ., 2022). In particular, field soil minirhizotron imagery has been analysed several times with good results (Wang et al ., 2019; Smith et al ., 2020; Gillert et al ., 2021; Han et al ., 2021; Peters et al ., 2022; Bauer et al ., 2022), but transferability between sites and out of agricultural soils is difficult to assess without widespread adoption in new settings. These have also never been applied to high frequency studies where variability between images (e.g.…”