Field Image Processing Pipeline Requirements

Goal and Objectives

The primary goal is to extract detailed vegetation segments from field images. These images include additional tools like gray mats and color cards to enhance the accuracy of segmentation. The segments extracted are intended to be used directly for detailed analysis or to create synthetic image data, facilitating diverse machine learning tasks.

1. Explore the dataset

   • check the status of the backlog

   • clean and organize backlogged data, understand it deficiencies, and review issues with current collection and data upload

   • output: organize the contents of various tables and blob containers into a single working table (.csv).

   • output 2: generate report of missing data, current status of the dataset and collection across locations, species, and plant types. Generate report about an major issues like the discrepency between uploaded jpgs, raws, and table entries.

2. Organize data into Batches for preprocessing

   • organzie data by state, capture date, and 3 hour time intervals

3. Develop Pipeline roadmap

   • outline image processing path

   • Semi-Automatic Labeling Techniques

   • Utilize a combination of classical digital image processing, and deep learning to generate multiple types of labels for the images, such as bounding boxes and segmentation masks. These labels are critical for training machine learning models and for validating the accuracy of computer vision algorithms.

4. Automate

5. Document

Key Features

Validation

\uD83E\uDD14 Assumptions

\uD83C\uDF1F Milestones

\uD83D\uDDD2 Outputs

 Open Questions

 Out of Scope