This is a project that has been on my mind for about a year now. Chalk this up as another crazy idea that could make someone a lot of money.
About a year ago, it was announced that K-State University (proud alumnus) would be partnering with Australia’s Queensland University of Technology in a really cool effort to use unmanned aerial systems (UASs) to detect pestilence. You can read the press article here: http://www.k-state.edu/today/announcement.php?id=18855
You are probably thinking that’s great Tony how can GPUs, graphical processing units, help with this. That a good question. About a month after this was announced from K-State, was the GPU Tech Conference, where Andrew NG was discussing using deep learning to distinguish pictures of mugs and cats. (http://www.gputechconf.com/highlights/2015-replays).
That’s when I started building on that idea in my head. Here is K-State collecting tons of data on what pestilence looks like from UASs and we are on the cusp of commercializing GPU based deep learning. What if we could do deep learning on all the material that was gathered from the initial study. Then leverage real time or near real time data streams from UASs to detect not only pestilence but all sorts of other agricultural ailments.
This has the potential to exponentially improve the agriculture (and many other industries as well). No longer will it be necessary to treat an entire field for a blight it can be done with precision reducing food costs and increasing productive time for the stewards of the land, farmers.
You may be thinking this is a niche market. This in fact could impact roughly 7 billion people. I don’t know about you but I like to eat, and because I don’t have the space in my back yard to be self sustaining I’m going to be dependent on vegetable farmers, grain farmers, citrus farmers, ranchers, and many other agricultural industries to make food stuffs available for me to purchase either directly or indirectly. The chart below from the USDA shows how the prices for various agricultural products have changed over the last 30 years.
Wouldn’t it be great if we could leverage technology we already have to help keep food bills lower? I suggest working on a project that leverages GPU and deep learning. How do we do it? I’m happy to make any introductions either in industry or at K-State.
Here is what it would take.
- HPC cluster with NVIDIA GRID GPUs
- A moderately sized SAN to store the resulting data
- Financial backing
- Integration with the UAS teams collecting the data
- A field to fly over
- Development of systems to feed data back for processing
- Technologist to develop the infrastructure
- Programmers to develop new learning techniques
The first 3 bullet points have to come from the technology industry. The last 5 points can be driven at the university level with guidance from the industries. You’ll probably notice I said industries… While the technology industry can help create this new paradigm it will need help from agricultural engineers and many others to make it rugged and viable.
I envision a high level design like this:
I don’t know a university that wouldn’t be interested in this project. If you’re in industry (ag, tech, engineering, or other) and are interested in helping with any of the 3 items in the list above let me know.
That’s all for now,
The Wondernerd
