About Platform 1

The first phase of the P2IRC program resulted in the development of numerous software applications, several of which are used in the image-processing pipeline. Others are used for specific domains such as the visualization of genomics data, soil analytics or microbiome workflows. These developments will be used across the program’s second phase and must be kept running, made more accessible and updated with new functionalities and other improvements as even more progress is made.

The Maintenance Platform focuses on three maintenance systems:

  • The PlotVision image-analysis pipeline;
  • The Deep Plant Phenomics platform; and
  • Several stand-alone applications for genome visualization, image exploration/annotation and data analysis.

Platform 1 will maintain, expand and advance these innovative agtech systems to improve their stability, usability, throughput and capability. It will also transition new software applications from testing environments to permanent hosts for general use. The aim is to - to effectively support the P2IRC program’s Flagship Projects and deliver the smooth interface between agtech input and innovative digital agriculture outcomes.

Practical Applications

  • PlotVision, the primary tool in the P2IRC program for managing and analysing image-based phenotyping data, allows researchers to browse, compare and assess imagery for crop field trials.
  • Each domain-specific application provides services to P2IRC and external researchers, including specific data analyses and data visualization.

The following projects are currently underway within Platform 1:

Activity 1.1

Maintenance of PlotVision

Maintaining and expanding the PlotVision system for the analysis of field images and the extraction of phenotypes.

PlotVision is a software-as-a-service product for the analysis of agricultural field images. It identifies individual field plots in drone images and uses machine learning to extract phenotypes (traits) and predict outcomes such as harvest yield and yield stability. P2IRC platform researchers will maintain the PlotVision system during this Activity, introducing upgrades and improvements as needed.

Researchers have recorded the following achievements throughout this Activity:

  • Maintained and updated the PlotVision database and server, including code improvement and testing, usability tests and the connection of provenance prototypes to existing pipeline systems allowing the processing of backlog data. The database now holds more than 40 trials from P2IRC researchers.
  • Improved the efficiency of several modules (upload, orthomosaic creation, image analysis) and integrated them with standard P2IRC workflows.
  • Updated several PlotVision tools to work with new image and sensor types, and several new analysis capabilities have been introduced.
  • Developed a data standard for automated field-book extraction that applies to all PlotVision modules.

Researchers have completed the conversion of other P2IRC data formats to the PlotVision standard.


Activity 1.2

Maintenance of deep plant phenomics

Maintenance of the Deep Plant Phenomics pipeline for automated phenotyping, and the integration of new functionalities.

Deep Plant Phenomics is an open-source deep learning tool that provides pre-trained neural networks for several common plant phenotyping tasks, as well as an accessible platform that can be used by plant scientists to train models for their own phenotyping applications. Platform researchers are maintaining the Deep Plant Phenomics platform, introducing new capabilities and ensuring its integration with PlotVision. The result will be a smooth interface between the imaging and sensing systems, combined to deliver extensive analyses of plant traits – allowing for informed decision-making in agriculture.

Platform researchers have successfully achieved the following:

  • Maintained and updated the Deep Plant Phenomics platform by incorporating an updated data storage framework from TensorFlow and upgrades for enhanced testing and interoperability.
  • Supported plot optimization methods Flagship Project 3 by incorporating new deep learning methods, including encoder-decoder architectures for image segmentation and two object-detection architectures.
  • Developed application programming interfaces to integrate Deep Plant Phenomics with PlotVision.
  • Begun work on the integration of plot optimization methods to accelerate the identification of individual plots in drone images, which is one of the current bottlenecks in image processing.

All new functionalities in the Deep Plant Phenomics platform have been archived in the GitHub repository.

Activity 1.3

Domain-specific applications

Enabling the transfer of domain-specific applications to a permanent host for general use.

The P2IRC program has led to the development of multiple software programs and algorithms currently restricted to domain-specific applications reflecting the initial programing environment. Through this Activity, researchers are working to make these applications ready for general use by transferring them to the program’s servers and improving their usability.

Three applications from P2IRC’s Phase I have updated to improve robustness, efficiency and usability (SynVisio, AccuSyn and Winnowing) and are now hosted on P2IRC servers. Initial work has been conducted to develop a procedure that brings new applications onboard for P2IRC researchers. Researchers have also introduced a load-balancing architecture onto the P2IRC web servers using NGINX, allowing data and computation to reside on machines other than the application host.