Five Insights from John Deere's Virtual Agronomy Summit on See & Spray Technology
A breakdown of John Deere's Virtual Agronomy Summit
Last week, John Deere hosted a Virtual Agronomy Summit spotlighting its See & Spray system.
The event offered a glimpse into the lessons learned from real-world usage and the insights driving this product's evolution.
From technical adjustments to practical applications in the field, to the potential of See & Spray in enhancing efficiency, and outcomes for growers, the Summit was a great breakdown of the See & Spray system.
In this summary, I’ll unpack some key takeaways from the Summit and what they mean for the future of agronomy and crop protection.
Index
The Need for Speed: Ultimate and 15mph Travel Speed and Rear Facing Nozzle Selection
Sensitivity Settings and Farmer Control: John Deere is Putting the “Egg” Back into the Product
Field Analyzer, Maps and Display: How are they determined?
71.5% of weed pressure is in the headlands and/or along waterways: Big Picture Implications for Crop Input Companies
Crop Injury Quantified
Final Thoughts
Note: All images below are courtesy of and sourced from John Deere with permission to publish unless otherwise stated.
1. The Need for Speed
To kick off the event, John Deere shared a nice comparison chart, illustrating the differences between See & Spray Ultimate and See & Spray Premium:
The biggest insight from this chart that I missed until this point is the speed that the “Ultimate” system can go: 15mph.
Source: Upstream Ag Insights
As I have talked about before, the constraint on speed stems from the needs of the system to reduce the time it takes for the end-to-end process: capture the image, process it through the model, make a decision, tell the right nozzle(s) to turn on then off. Some of the time comes from machine learning processing, and some comes from the communication (getting an image from the camera to the controller and sending the message to the nozzle).
Tough to execute at higher speeds and maintain adequate control.
The bottle neck to the above then is inference latency (eg: how long the neural network takes to determine if a weed is in the image). If a company can reduce this latency without increasing detection error, a farmer can drive faster.
Reducing latency of any networked process comes from two things:
faster software (algorithm improvements/optimizations)
faster hardware (eg: microchips)
I am not certain what improvement capability would come from each, but I suspect the biggest contributor is hardware (80%+). Which means a company like John Deere is reliant on chip improvements from NVIDIA (note: there may be some component of camera position to give incremental lead time, too).
What I did not consider is nozzle selection.
John Deere talked about how rear facing nozzles help with speed they can go.
15mph is still slower than many broadcast applications will go (17+mph), but that gets incrementally closer to the same sort of in-field efficiency as a broadcast application.
Next, Deere shared some insights into capability by crop:
Along with the different tank modes they have:
The dual independent tank is where See & Spray gets incredibly compelling as more than a herbicide tool and becomes a multi-use agronomic solution— enabling broadcast sprays of fungicides, nutrients or biostimulants for example while also being able to manage weeds. This doesn’t even include the data insights acquired from the cameras on the system being captured every pass!
2. Sensitivity Settings
John Deere shared the ability to set the sensitivity of the system to focus on control or savings:
There is a nice comparison that was shared to illustrate low vs. medium vs. high (only high and low shared by me, Deere did share medium, too):