bidopsis thaliana data set ( https://zenodo.org/record/50831#.XjIAPVNKhQI ), wheat seedling data set ( http://gigadb.org/dataset/100346 ), and barley data set ( https://www.quantitative‐plant.org/dataset/3d‐magnetic‐resonance‐images‐of‐barley‐roots ). The source code and pre‐trained SR models are available at GitHub and Zenodo (https://github.com/GatorSense/SRrootimaging; https://doi.org/10.5281/zenodo.3940562 ; Ruiz‐Munoz, 2020 ). LITERATURE CITED Akinnifesi , F. K. , B. T. Kang , and D. O. Ladipo . 1998 Structural root form and fine root distribution of some woody species evaluated for agroforestry systems . Agroforestry Systems 42 : 121 – 138 . Araus , J. L. , and J. E. Cairns . 2014 Fiel
Open resource ↗GatorSense/SRrootimaging · lines:155-276Unverified paper record
Super resolution for root imaging.
Applications in Plant Sciences · 1 Jul 2020 · 10.1002/aps3.11374
Abstract
PREMISE: High-resolution cameras are very helpful for plant phenotyping as their images enable tasks such as target vs. background discrimination and the measurement and analysis of fine above-ground plant attributes. However, the acquisition of high-resolution images of plant roots is more challenging than above-ground data collection. An effective super-resolution (SR) algorithm is therefore needed for overcoming the resolution limitations of sensors, reducing storage space requirements, and boosting the performance of subsequent analyses. METHODS: We propose an SR framework for enhancing images of plant roots using convolutional neural networks. We compare three alternatives for training the SR model: (i) training with non-plant-root images, (ii) training with plant-root images, and (iii) pretraining the model with non-plant-root images and fine-tuning with plant-root images. The architectures of the SR models were based on two state-of-the-art deep learning approaches: a fast SR convolutional neural network and an SR generative adversarial network. RESULTS: In our experiments, we observed that the SR models improved the quality of low-resolution images of plant roots in an unseen data set in terms of the signal-to-noise ratio. We used a collection of publicly available data sets to demonstrate that the SR models outperform the basic bicubic interpolation, even when trained with non-root data sets. DISCUSSION: The incorporation of a deep learning-based SR model in the imaging process enhances the quality of low-resolution images of plant roots. We demonstrate that SR preprocessing boosts the performance of a machine learning system trained to separate plant roots from their background. Our segmentation experiments also show that high performance on this task can be achieved independently of the signal-to-noise ratio. We therefore conclude that the quality of the image enhancement depends on the desired application.
Plant phenotyping relevance
植物根の低解像度画像を高解像度化する深層学習手法を開発・比較し、根画像の分割性能への効果も検証しており、表現型取得・抽出法が中心である。
abstractAn effective super-resolution (SR) algorithm is therefore needed for overcoming the resolution limitations of sensors
abstractWe propose an SR framework for enhancing images of plant roots using convolutional neural networks.
abstractWe compare three alternatives for training the SR model
abstractWe demonstrate that SR preprocessing boosts the performance of a machine learning system trained to separate plant roots from their background.
Code and data availability
The paper's authors explicitly state that their source code and pre-trained SR models are publicly available on GitHub and Zenodo, and they used five publicly available image datasets (DIV2K, 91-Image, Arabidopsis thaliana root data, wheat seedling roots, 3D MRI barley roots) plus a SegRoot soybean test set as phenotyp
WinRHIZO images of the barley roots. In our experiments, we grouped the three plant‐root data sets into a single data set named “Roots.” Figure 3 shows examples of the plant‐root data sets used for training the SR model. To test the performance of the SR models, we used a data set of 65 soybean ( Glycine max (L.) Merr.) roots ( https://github.com/wtwtwt0330/SegRoot [accessed 11 June 2020]) (Wang et al., 2019 ). SR model training Many CNN architectures that enable the mapping of LR images into SR images can be found in the machine learning literature. In this study, we used two state‐of‐the‐art CNN‐based models, FSRCNN and SRGAN, to convert LR root images to SR images. FSRCNN is a model th
Open resource ↗wtwtwt0330/SegRoot · lines:93-102In this study, we used five publicly available data sets to train the SR models. We used two non‐plant‐root data sets, DIV2K ( https://data.vision.ee.ethz.ch/cvl/DIV2K/ [accessed 11 June 2020]) and 91‐Image ( https://www.kaggle.com/ll01dm/t91‐image‐dataset [accessed 11 June 2020]). DIV2K is a data set of natural images that has been used by others to train and test SR algorithms (Timofte et al., 2017 ). We trained our models on the grayscale version of this training data set (800 images). The
Open resource ↗lines:93-102In this study, we used five publicly available data sets to train the SR models. We used two non‐plant‐root data sets, DIV2K ( https://data.vision.ee.ethz.ch/cvl/DIV2K/ [accessed 11 June 2020]) and 91‐Image ( https://www.kaggle.com/ll01dm/t91‐image‐dataset [accessed 11 June 2020]). DIV2K is a data set of natural images that has been used by others to train and test SR algorithms (Timofte et al., 2017 ). We trained our models on the grayscale version of this training data set (800 images). The 91‐Image information is a classical data set commonly used in SR studies. We also used t
Open resource ↗lines:93-102This is an automatically classified, unverified record. Curator approval is required before any resource enters the Catalog.