In the collaboration with Honaas lab at the USDA Agricultural Research Service, the Ficklin Research Program at Washington State University has developed Granny, an image analysis software that uses an instance segmentation algorithm to identify individual fruit in photos, and then process the images to extract individual fruit sub-images and remove the background for downstream analyses. Granny, aiming to assist in post-harvest fruit maturity index experiments, generates useful results from RGB images for determining superficial disorders and maturity in many apple and pear cultivars. Granny is a computer vision software aiming to assist technicians in post-harvest fruit maturity index experiments.
- Nhan Nguyen - Ficklin lab's undergraduate researcher
- Joseph Michaud - Biological Science Technician, USDA-ARS Hood River
- Heidi Hargarten - Honaas lab's postdoctoral researcher, USDA-ARS Wenatchee
- Loren Honaas - Honaas lab's principal investigator
- Stephen Ficklin - Ficklin lab's principal investigator
Development of Granny was funded by the USDA Agricultural Research Service and the Washington Tree Fruit Research Commission under project AP-22-101A.
The main usage of the program consists of 2 main steps:
- Step 1. Locate and extract instances using Mask-RCNN [Instance Extraction].
- Step 2. Rate each instance. Depending on your purpose, you can tell the program to
- rate superficial scald in Granny Smith apples [Superficial Scald]
- calculate starch area on iodine-stained cross-sections [Cross-section Starch]
- sort color of the pears [Pear Color]
Please refer to our demo page for a walk-through of each step's input and output.
Please refer to our docs page for a detailed explanation of how to install the program and related packages in Python.
We are sorry in advance for the inconvenience, but our program still contains a few limitations.
- [Instance Extraction] Due to the dependency requirements, it is recommended to install Python's package managers such as conda or virtualenv. Here, we provide instructions for installation of packages using conda.
- [Superficial Scald] Due to the similarity in coloration, the stem of the apples could be potentially counted towards the total area of superficial scald.
- [Cross-section Starch] In order to run our provided ImageJ scripts for starch analysis, the user must seperately install Fiji. We are currently trying to implement the Python wrapper function for calling ImageJ macros.
- [Cross-section Starch] Due to the similarity in coloration, the code is unable to distinguish between bruising and iodine-stained area.
- [Pear Color] Currently, the color of single pear image is only being sorted based on the provided color card under demo/pear_images/color_preference.