Real-time Seed Identification

Name: Dan Curry
Affiliation: Crop and Soil Science Department, OSU
Phone: 541 207-4247
E-mail: daniel.curry@oregonstate.edu
Website: http://seedlab.oregonstate.edu/
Knowledge Required: Must be proficient in the necessary programming languages, able to put components together to develop a working model that delivers a yes/no decision on the identity of each seed (yes it is a seed that is like the 99% in the sample or no, it is not).

Motivation: Currently, OSU seed analysts observe 25,000 grass seeds for each sample, pulling out by hand all of the off-types. A seed analyst may perform this on ten different samples per day, which causes eyestrain, headaches, backaches and tends to be boring. Currently there are machines that can identify larger seeds, like cereals (wheat, oats and rice) and separate the off-types. The challenge is that the grass seed market is a bit smaller so no one has developed a machine. The machine developed with this project could be the first to do so for grass seed, which would make the seed analysts more efficient in their jobs. If done correctly, it could also improve the accuracy of the final report and change the seed testing industry.
Description: Seed Identification using computer vision, machine learning, Jetson TX2 processor and high a resolution camera to identify seed and output a signal.

Objectives: To use the latest equipment available to develop a machine that will enhance the work performed by seed analysts.
Deliverables: A working machine that can identify each seed and separate the pure seed from weed seeds, other crop seeds and inert material by sending a signal. A mechanical separator that accepts the yes/no signal will be developed by others.

Other comments: Feel free to call me at my office on campus: 541 737-5094 or by email: Daniel.curry@oregonstate.edu.


   D. Kevin McGrath
   Last modified: Fri Oct 20 09:31:13 2017