SOW_seeotter_consultant_2.docx

STATEMENT OF WORK

1.0 INTRODUCTION

Contractor is to provide expertise and consultation to expand accessibility, repeatability, and documentation capabilities of SeeOtter, an AI-assisted, wildlife photo processing application. SeeOtter is currently being used for sea otter aerial photographic surveys in Alaska. Improvements will expand the longevity and effectiveness of the coding and models for future surveys and new users.

2.0 BACKGROUND

Aerial surveys are a common tool used to monitor status of wildlife populations. US Geological Survey, US Fish and Wildlife (USFWS), and National Park Service (NPS) have a shared interest in the status and population trends of sea otters across Alaska. Historically, sea otters in Alaska were counted by observers in small airplanes flying at a low altitude and low airspeed. Given the many risks involved with flying low and slow for long periods over water, we’ve spent the past six years developing methods to transition from observer-based to photography-based aerial surveys, which allow for higher altitude flying and increased safety. Although photographs allow for permanent photographic record of sea otters, using photography introduces new challenges related to data storage and image processing.

After years of manually processing photos by a human observer from small survey areas, two contractors hired by USFWS developed an AI-assisted model, SeeOtter, that processes images using Python-based coding with a user-friendly interface. Our goal is to decrease time spent organizing, deleting, cropping, and processing images. SeeOtter consists of a YOLO v5 model to detect potential sea otters in images, and the viewer can mark potential sea otters as yes, no, or ambiguous. Ambiguous tags are then checked by independent reviewers. All these features make processing images more time-efficient and consistent for a broader working group.

3.0 SCOPE

The purpose of this project is to update and improve SeeOtter modeling and code efficiency so that USGS can continue to increase accuracy and speed of sea otter image processing after photo-based surveys.

5.0 TECHNICAL REQUIREMENTS

· Integrating current python codes into SeeOtter: Currently there are several pre-process steps that occur. We want to streamline these steps into a more user-friendly interface.

· AI Model Compatibility Expansion: enhance SeeOtter's adaptability by enabling compatibility with a variety of object recognition models. Establish a flexible framework within SeeOtter that can seamlessly integrate various object recognition models.

· Increasing SeeOtter functionality: incorporate additional photo sensors into SeeOtter. Develop a more flexible image processing pipeline within SeeOtter.

· Development of In-Depth Guidance Materials: Development of comprehensive openly available documentation for SeeOtter includes several key components and adheres to the open data and open science mandate of DOI.

· Workflow Development for High-Power Computing: establish a cloud-based data management plan, allowing researchers to efficiently manage, store, and access large datasets remotely through Caldera, USGS’s high performance data storage.

Criteria for acceptance of Product or Service (how will contractor's compliance be measured).

The specific work requirements (above) will be performed collaboratively with ASC biologists. Compliance to the stated needs from the contractor will be evident in their interactions. Also, the defined deliverables (below) will be assessed to ensure that they comply with the stated needs from the contractor.

6.0 DELIVERABLES

1. Characteristics for work products described in 5.0 Technical Requirements, identified by name and quantity. The following will be delivered directly to the USGS technical contact (to be provided with the award):

a. SeeOtter coding files

b. User manual for applying new models through Python coding in SeeOtter

c. Updated user manual for SeeOtter

d. HPC user instructions for access and job processing

2. Delivery dates for work/product to be completed and point of delivery

a. Work will be completed by Sept. 30, 2024, with delivery to and compliance checks by ASC biologist.