Biology subjects

Pound, M.

Publications and source records attributed to Pound, M..

3 recordsLinked to original sources

bioRxiv2021

CNN based Heuristic Function for A* Pathfinding Algorithm: Using Spatial Vector Data to Reconstruct Smooth and Natural Looking Plant Roots

In this work we propose an extension to recent methods for the reconstruction of root architectures in 2-dimensions. Recent methods for the automatic root analysis have proposed deep learned segmentation of root images followed by path finding such as Dijkstras algorithm to reconstruct root topology. These approaches assume that roots are separate, and that a shortest path within the image foreground represents a reliable reconstruction of the underlying root structure. This approach is prone to error where roots grow in close proximity, with path finding algorithms prone to taking "short cuts" and overlapping much of the root material. Here we extend these methods to also consider root angle, allowing a more informed shortest path search that disambiguates roots growing close together. We adapt a CNN architecture to also predict the angle of root material at each foreground position, and utilise this additional information within shortest path searchers to improve root reconstruction. Our results show an improved ability to separate clustered roots.

plant biology↗

bioRxiv2020

PhenomNet: Bridging Phenotype-Genotype Gap: A CNN-LSTM Based Automatic Plant Root Anatomization System.

This research will explore the phenotype-genotype gap by bringing two very diverse technologies together to predict plant characteristics. Currently, there are several studies and tools available for plant phenotype and genotype analysis. However, there is no existing single system that offers both capabilities in one package. Usually, Convolution Neural Networks used for plant phenotyping analysis and Recurrent Neural Networks used for genotype analysis. Both of these machine leanring methods require different input data for feature extraction, analysis and learning. Building a machine learning system for plant data that can make use of both graphic (for phenotype) and time-series (for genotype) is critical and challenging, especially when the system has to predict sensitive information regarding plant growth, accession and types. In this study, the proposed system will solve these problems by bringing two very different technologies, analysis methods and datasets. The proposed research aims to bridge the phenotype-genotype gap using CNN-LSTMs to process graphic and temporal data of plant roots. The proposed system "PhenomNet" offers segmentation of plant roots along with the classification of the given dataset into different accessions. The experiment results have shown that proposed CNN-LSTM architecture provides very high accuracy in comparison to manual or semi-automated approaches.

plant biology↗

bioRxiv2020

RootNet: A Convolutional Neural Networks for Complex Plant Root Phenotyping from High-Definition Datasets

Plant phenotyping using machine learning and computer vision approaches is a challenging task. Deep learning-based systems for plant phenotyping is more efficient for measuring different plant traits for diverse genetic discoveries compared to the traditional image-based phenotyping approaches. Plant biologists have recently demanded more reliable and accurate image-based phenotyping systems for assessing various features of plants and crops. The core of these image-based phenotyping systems is structural classification and features segmentation. Deep learning-based systems, however, have shown outstanding results in extracting very complicated features and structures of above-ground plants. Nevertheless, the below-ground part of the plant is usually more complicated to analyze due to its complex arrangement and distorted appearance. We proposed a deep convolutional neural networks (CNN) model named "RootNet" that detects and pixel-wise segments plant roots features. The feature of the proposed method is detection and segmentation of very thin (1-3 pixels wide roots). The proposed approach segment high definition images without significantly sacrificing pixel density, it leads to more accurate root type detection and segmentation results. It is hard to train CNNs with high definition images due to GPU memory limitations. The proposed patch-based CNN training setup makes use of the entire image (with maximum pixel desisity) to recognize and segment give root system efficiently. We have used wheat (Triticum aestivum L.) seedlings dataset, which consists of wheat roots grown in visible pouches. The proposed system segments are given root systems and save it to the Root System Markup Language (RSML) for future analysis. RootNet trained on the dataset mentioned above along with popular semantic segmentation architectures, and it achieved a benchmark accuracy.

plant biology↗