Specimen-GT tool: Ground Truth Annotation tool for herbarium Specimen images

Digitized herbarium specimens are very important for scientists belonging to medical, biological, chemical, botanical and in general bio scientist’s communities. Extracting traits and identifying species of plants are highly demanded even for species inventory and traits extraction.

Information on specimen are considered as metadata and can be then stored in a standard format (ie ABCD). As for the specimen images content, it needs tools and techniques to make it available for scientist especially the morphological traits (https://www.try-db.org). Measuring leaves for example need at first use of segmentation techniques before the measurement process.

Recently, with the success demonstrated by machine learning techniques to solve segmentation challenges there is a growing need for ground truth data not only for the purpose of evaluation but also for labelling learning datasets. The labelled datasets are then stored in standard and structured format linking data and metadata.

Specimen-GT tool is a system for editing and reviewing specimen image data and metadata. It provides the following functionalities:

1. Generate data structure to store and manage ‘ground truth’ data.
2. Generate and respect naming convention between specimen images data and annotation files
3. Provide a tool box to extract measurement from specimen leaves.
4. Interact with Bexis platform
5. Organize annotation data and metadata into files according to a dynamic structure:
6. Generate file SHAPE-data
7. Generate Content annotated file DATA.
8. Generate ABCD-data to manage store and share meta-data
9. Generate resized specimen image data
10. Visualize annotated data from specimen
11. Provide a toolbox to customize (add-update-delete) attributes and associated types (shape, number, text, etc).

Actually, the tool is used to create a full annotated dataset with (leaves measurements, missed part leaf, perfect leaf, herb, overlapped leaves). The annotated images count over 4000.