Image/Data Analysis

Recorded images are processed to extract information from the image data via computer vision or machine learning procedures. The parameters extracted from the images serve to generate knowledge on the analyzed sample. Regardless of what imaging principle is used, the processing serves to find the target objects in the images and to measure properties of these target objects. In the example, an RGB image is processed to find the target object, a plant, separated from the background, pot, or carrier. The detected plant is measured for size, shape, and color.

Self-Programmed Analysis of Images with LemnaGrid

LemnaGrid is the most flexible phenotyping software for user programming. The LemnaGrid software module uses an intuitive graphical programming environment, which allows easy integration of different image analysis algorithms. The image processing workflow extracts desired properties/features from the original image and stores results in a dedicated storage system. LemnaGrid can include functions of state-of-the-art machine learning tools, next to classical image analysis to detect and extract features of interest. Guided by the LemnaGrid graphical user interface, users can build their own analytical workflows with pre-defined functions. LemnaTec provides a comprehensive user-programmable toolbox for image and data processing, the LemnaGrid package. With this package users can program image processing workflows without the need for command line programming. LemnaGrid provides a broad library of functions that are required for image processing. Functions comprise image handling, image analysis, and data output.

Analysis Tools Include

  • filters

  • threshold functions
  • image converters
  • object detectors
  • feature detectors
  • color analysis tools
  • region of interest functions
  • combinatory functions

All functions – graphically represented by boxes – can be combined on a surface by aligning the boxes to workflows. Along such workflows, each function can provide insight in the progress of the analysis by feeding out an intermediate image. Image processing workflows can include branches and parallel paths that combine to a final result. LemnaGrid extracts a wide selection of relevant features out of images recorded with different sensors and derives phenotypic parameters relevant in plant research, breeding, environmental studies, seed testing, toxicology, and more. Besides RGB, LemnaGrid can process images coming from many camera types such as chlorophyll fluorescence, hyperspectral, multispectral, or (near-)infrared cameras. Moreover, images of luminescent or fluorescent reporters in genetic and physiological assessments can be processed. LemnaGrid analytical software enables extraction of information from all recorded images and scans. The image and scan data, together with comprehensive metadata, are stored into a file system or data base, and are accessible for retrieval for analysis. LemnaTec software always stores the unprocessed original data coming from the cameras, so that users can access these for documentation, quality control, or re-assessments. All subsequent image processing is documented, so that all steps of the analyses are transparent.

Customized Analysis of Images and Machine Learning

For complex analytical tasks, LemnaTec provides application-specific and customized analyses; many of them include deep learning. Machine learning algorithms allow for training the analysis system to customer-specific parameters. Algorithms are trained to specifically recognize features in the images and use these for classification.

How it Works

In a first step, a series of typical images of the samples are recorded and target features are labeled using the LemnaTec Labeling Tool. Thereafter, labeled images are used to train the algorithms. In serial training iterations, models are adapted to recognize the target features in the images. Once the model works satisfactory, it can be applied to images coming from the actual testing process, i.e., the experiment or the quality assessment.

Example with Germinating Seeds

In this example with germinating seeds, the model was trained to detect seeds, roots, and shoots versus the background. Each detected target feature is colored with a specific label. Here, yellow denotes the seed grains, brown stands for the roots, and magenta for shoots. Any detected target object is not only counted, but it is measured for parameters such as length, width, or area. All results of the analyses are provided in the LemnaAIxplorer module.

Good To Know

  • Computer Vision or Machine Vision: classical image processing without learning algorithms
  • Machine Learning: analyses based on learning algorithms
  • Deep Learning: analyses using artificial neuronal networks
  • Artificial Intelligence: computer-assisted analyses

Features Assessed in Processing of RGB Images

RGB images are used for a broad range of applications like

  • Your Content Goes Here
  • area
  • length, width, height
  • counting
  • compactness, roundness, eccentricity, convex hull
  • ratios of measured parameters
  • architectural and skeleton factors
  • color-related information such as greenness

Multi-view RGB images can be used to calculate
3D representations of the parameters.

Using non-RGB data increasingly gains importance for physiological phenotyping. LemnaGrid can process images achieved with fluorescence, luminescence, (near-) infrared, hyper- and multispectral cameras as well as 3D imaging systems.

Features Assessed in Processing of Fluorescence Images

For fluorescence imaging, occurrence, localization, and intensity of the fluorescence signals – autofluorescence and fluorescent biomarkers – are analyzed. For fluorescence measured with Kautsky or PAM protocols, a broad range of standard chlorophyll fluorescence parameters describing the status and activity of the photosystem II are provided.

Features Assessed in Processing of Bioluminescence Images

High light sensitivity imaging enables luminescence recording, and resulting bioluminescence images, containing signals from luminescent biomarker, or labelled microbes are processed for intensity and localization of the luminescence.

Features Assessed in Processing of Infrared Images

Infrared images are analyzed for intensity of the IR radiation and the spatial distribution of the corresponding temperatures. Images of different sources can be overlaid to co-localize morphological features from RGB-images with e.g., luminescence or fluorescence.

Features Assessed in Processing of Hyperspectral Images

Hyperspectral images reveal spectral reflectance data that are localized on the plant surface, and can be processed for physiological information, once correlated to the plant’s physiological properties. More than 60 vegetation indices are pre-set in the LemnaTec software, including NDVI, OSAVI, MCARI, NPCI, NPQI, SIPI, PRI, Anthocyanin, Carotenoid, Lichtentaler, Vogelmann, Carter, Disease indices, Water-related indices, and more.

Features Assessed in Processing of 3D Laser Scans

For advanced structural imaging, laser scanners are used, and the recorded point clouds deliver information on spatial data of plant structure, dimensions, and on angles between plant parts.