Pre-trained deep learning model for vineyards?

[JimWooten]

Hi,

Does ESRI have a pre-trained deep learning model for vineyards? Alternatively, is it possible to use a text SAM to detect vineyards from satellite imagery?  Thanks.

Jim

[NickShort1 · Accepted solution]

Hi Jim,

We don't have any pre-trained models for vineyard detection or analysis.   In reading Vineyard Zoning and Vine Detection Using Machine Learning in Unmanned Aerial Vehicle Imagery,  it seems that our toolset could be applied, but the resolution requirements suggested in the paper are for < 25cm.   That might be too costly with satellite data.   

 

[JimWooten]

Thanks, Nick, and I'm quite sure you're correct about the cost. The project was to be for a vineyard association and they don't have a budget for the high-resolution satellite imagery.  And as this is for detection of unregistered vineyards, they don't want to fly UAVs over unknown property. I'll try the Prompt Based Segmentation model and see what I can do.

Thanks for your reply.  I'll accept as a solution but it would be nice to have a model that does this in case anyone is asking. It's certainly an important industry in California and growing in other states.  Thanks again.

Jim

[NickShort1]

Jim,

One thought might be to use free NAIP imagery as it has the spatial resolution, though it only goes over the same area every 2-3 years.
Since these wineries don't change that often, that might satisfy your requirements.
We've built some ML models against orange groves in NAIP, which is covered in: https://www.researchgate.net/publication/349462747_Precision_agriculture_and_geospatial_techniques_f...
While not the same problem, it may be possible to apply similar techniques to vineyards since there is a bit of structure to the fields. In otherwords,
train the ML models to do object detection using the known vineyards. Send me an email if you want to discuss further (nshort@esri.com),

[JimWooten]

From what I'm told, the wildcat vineyards are quickly developed from vines obtained illicitly, allowed to mature and produce grapes, and then plowed under.  Apparently, it's causing enough of a problem for the legally registered NJ vineyards to impact their profit margins. The NAIP imagery pass-over might not catch the illicit activity.

I'll send an email later today or first thing tomorrow.  Thanks.

[JimWooten]

It's a bit surprising to me that, given the importance of the wine industry to California's economy, and ESRI's HQ location, that Rohit Singh hasn't been tasked with developing a pre-trained model for vineyards.  According to Forbes: Nationally -  $323.55 billion; 1.75 million jobs; $102.14 billion in wages.