Rapid Agrichemical Inventory via Video Documentation and Large Language Model Identification

Area of Science:

• Environmental Science
• Occupational Health
• Agricultural Science

Background:

• Presents a novel methodological approach for agrichemical inventory documentation.
• Complements exposure assessments with time-restricted observational methods in field settings.
• Utilizes large language model (LLM) capabilities for agrichemical categorization.

Purpose of the Study:

• To develop and evaluate a rapid method for documenting agrichemical inventories using LLMs.
• To assess the feasibility of categorizing agrichemicals from video footage under time constraints.
• To enhance exposure assessment strategies in field research.

Main Methods:

• Recorded a short video of agrichemicals in a storage shed.
• Processed video into 31 screenshots for analysis.
• Employed OpenAI's ChatGPT (GPT-4o) for agrichemical identification and categorization.

Main Results:

• LLM accurately identified 75% of agrichemicals.
• Human verification was used to correct and validate LLM entries.
• Demonstrated feasibility of LLM application in time-limited field research.

Conclusions:

• The LLM-assisted method facilitates rapid initial data collection for exposure assessments.
• This approach is valuable in situations with limited researcher access to hazardous materials.
• LLM technology offers efficiency and cross-validation, enhancing field-based research capabilities.