Bayer AG utilizes Apps to train and deploy AI models for digital phenotyping

Overview

Plant phenotyping is key for automated and unbiased monitoring of plant heath in chemical screening or filed applications. Commonly used methods to tackle this include deep learning models for semantic segmentation as well as feature extractions. However, most of these software solutions require a deep understanding of coding and are not accessible to the daily user.

Therefore, by utilizing MATLAB App Designer, already existing neuronal networks trained in e.g. Python could be used in a simple one click solution and even training of new networks can be started via an intuitive graphical user interface. In addition, the graphical user interface can be used to compare different networks in terms of accuracy and complete analysis pipeline can be started with ease. 

Highlights

• Training of deep neural networks for semantic segmentation in MATLAB
• Import of pretrained networks from other Deep Learning frameworks
• Development of a custom App for others to train new models and compare results of different networks
• Integration of AI based algorithm in an analysis pipeline

About the Presenter

Martin Schmuck (Bayer AG)
Master of Science in Chemistry, University of Heidelberg, Germany (focus on physical chemistry)
Dr. rer. nat., University of Düsseldorf, Germany (High Content Image Analysis and software development)
Postdoc, UC Davis, USA, (High Content Image Analysis and software development)

Since 2018 Martin Schmuck works as an expert for screening technology at Bayer focusing on:

• Develop screening approaches for plant phenotyping and pharmaceutical research
• Develop custom tailored software solutions including deep learning methods
• Make deep learning available to untrained personal via easy-to-use Apps
• Deploy algorithms in cloud pipelines in production