Small Collections Grant
This page is used to provide assessment scores for each grant application assigned to you. Please use the rubric below the grant details to enter your assessment scores and any notes you wish to include.
Please note that all scores entered must be whole numbers (no decimals), or you will be unable to save this form.
The Relevance of a Small Amazonian Herbarium: The Use of Deep Machine Learning for Biodiversity Knowledge
Brazil
Digitally image
Cost (USD):
2000
HUAM
Universidade Federal do Amazonas
Objective:
This project leverages deep machine learning to enhance the scientific value of the Federal University of Amazonas (HUAM) herbarium. By analyzing a digitized collection of 12,000 herbarium specimens and incorporating online repositories, we aim to:
*Develop a machine learning model for automated and accurate species identification.
*Improve research and extension activities at HUAM by streamlining the identification process.
*Address possible misidentifications in tropical plant collections, contributing to broader knowledge advancement.
Timetable:
The project is currently progressing through four phases. In the initial two years, we initiated the digitization of the HUAM collection, and successfully processed 8,000 specimens. Simultaneously, we are actively downloading and curating well-identified images from online repositories to enrich our dataset.
The subsequent six months will be dedicated to initiating the training of the machine learning model. We will utilize existing images and curated datasets, supplemented by additional images from online repositories. The continuous refinement of the deep learning model will be a focal point, with rigorous testing and validation to ensure optimal performance.
Post-model training, a three-month phase integrates the model seamlessly into HUAM's workflow. Staff training ensures the smooth adoption of the automated identification system.
In the conclusive phase, the final three months will prioritize a comprehensive evaluation of the model's effectiveness. This involves assessing accuracy in comparison to manual identifications and addressing any potential challenges or improvements needed. Additionally, the phase encompasses documenting and disseminating the project findings.
Scoring Rubric
Reviewer's name:
Collection Improvement (max. 120 points)
-
Facilitating access to the physical collections by digitization (e.g., data entry, setting up database structure with an outline of the platform to be used, purchasing equipment, and imaging specimens) – up to 30 points.
-
Enhancing physical collections by improving the conservation status of specimens in the herbarium (e.g., better folders, protecting covers, mounting paper, labeling, etc.) – up to 30 points.
-
Curating specimens (e.g., updating families, species identification, identifying types) – up to 20 points.
-
Increasing our understanding of the flora or funga by making new herbarium specimens available, such as processing of backlog or collecting and mounting of new specimens from understudied sites – up to 20 points.
-
Securing collections by distribution of duplicates (or orphan collections) to other regional or international herbaria or shipping endangered collections to another herbarium – up to 20 points.
This proposal scores:
/120
Methods & Funding (max. 40 points)
-
Match between the proposed budget and methods for the aims described – up to 10 points.
-
Perceived need, the extent to which the project will benefit from IAPT funding: e.g., due to active floristic work or contribution to poorly collected sites, due to threatened conditions of collections, and for the degree of involvement of others (outreach and education). We give more points for herbaria in low- and middle-income countries – up to 20 points.
-
Sharing duplicate specimens with other herbaria – up to 10 points.
This proposal scores:
/40
Broader Impacts (max. 40 points)
-
Degree of regional importance of the collection or the taxonomic importance of the targeted collection – up to 10 points.
-
The project will yield durable benefits (specimens, digitized metadata, databases, websites) – up to 15 points.
-
The project involves outreach/mentoring and broad dissemination – up to 15 points.
This proposal scores:
/40
Year of last successful SCG application:
Has applicant applied for SCG before?:
Plan:
In my role as a member of the Herbarium at the Federal University of Amazonas (HUAM), I am conducting a collection-focused doctoral research within Bionorte's postgraduate program. My thesis, titled "The Relevance of a Small Amazonian Herbarium: Utilizing Deep Machine Learning for Biodiversity Knowledge".
This initiative involves employing deep learning algorithms to train a model for automated image classification to advance species identification. Rigorous testing and validating processes will guarantee accuracy, followed by staff training for seamless integration into HUAM's workflow.
The project encompasses the comprehensive digitization of the entire HUAM collection, incorporating 12,000 images from specimens. Additional images from online repositories of various herbaria will augment the machine learning model's training, featuring well-identified specimens. Remarkably, our herbarium already possesses 8,000 photographed species, serving as a foundational resource for this initiative.
Executing the machine learning models demands a high-performance computer, especially focusing on the graphics components. This underscores the imperative need for computing resources to support the computational demands of deep learning processes.
This strategic approach aims to address challenges in accurately identifying species, particularly within tropical plant collections, positioning HUAM as a pioneer in utilizing deep learning for biodiversity knowledge in the Amazon.
Institution:
IH Code:
Country:
Target areas:
Applicant First Name/s:
email:
"Other" target:
Deisy Pereira Saraiva
Applicant Last Name/s: