Cotton & Robotics: 3 Applications to Improve Production
October 1, 2020

Cotton Incorporated explores robotics’ potential.

Read this white paper to get a detailed account of five different research projects sponsored by Cotton Incorporated exploring robotics’ potential to improve cotton production and yields and farmer profits.

 

Rapid advances in autonomous equipment and machine learning are making widespread agricultural robotic use an imminent reality. Most of the major agricultural machinery companies have already announced plans for autonomous agricultural robots, and early prototypes have shown particular potential for weed control and crop harvesting.

 

Over the last two years, Cotton Incorporated has sponsored projects at seven universities within the Cotton Belt to evaluate the use of robotics platforms and identified weed control, machine vision systems and multiple harvesting as some of the highest potential impacts for production.

 

1.   Robotic Weed Control

 

Robotic weed control provides an elegant answer to two competing weed issues: herbicide-resistant weeds and the environmental impact of runoff. Robotic weed control can reduce the need for weed killers — eventually saving famers money and protecting long-term soil health — in two key ways:

 

  • Alternative weeding: Robotics can employ diode lasers and mechanical weeding as an alternative to herbicide. Laser weeding helps control weeds in the same row as cotton with minimal crop impact.
  • Targeted weeding: Machine learning coupled with machine vision technologies can better differentiate between weeds and cotton plants and identify the best method to eliminate a particular weed or spot-spray instead of blanket spray herbicide.

 

A Cotton Incorporated-sponsored project at Clemson produced a mechanical weeder prototype based on current commercial tillers, as well as an autonomous robotic platform that can be used to spray, weed or harvest.

GHG Protocol

2.  Machine Vision Systems

 

Training artificial intelligence to identify weeds and cotton bolls at the correct harvesting levels is by far one of the largest impediments to robotic agriculture because of the labor involved in sourcing, labelling and uploading the large number of images necessary. Open-source libraries have provided a key answer in both areas for Cotton-sponsored researchers at the University of Georgia.

 

  • Harvesting: By augmenting their 2085 cotton boll images with 56,295 images from an open source library they trained an AI system in only 4 hours. In the resulting field tests, their autonomous robot was able to successfully pick 77% of the cotton bolls it encountered.
  • Weeding: UGA also trained their system to identify 12 weed species by adding 21,114 images to their 782 from an open-source library. Their model has been able to identify images in the database with 86% accuracy.

Hardware costs for the UGA machine vision system (camera, processor, software) were under $1,000. Cotton Incorporated’s support for image libraries on cotton parts and cotton weeds, might help this system soon become a viable solution for cotton farmers.

3.  Multiple Harvesting

 

With rainfall and tropical storms on the rise in much of the Cotton Belt, the current, single-pass mechanical harvesting system often leads to harvest delays. Even under the best-case scenarios, mature cotton bolls positioned lower part of the plant are left exposed to weathering for over 50 days while the bolls positioned higher mature.

 

Robotics makes the option for multiple harvests—when each boll opens—a possibility, and Cotton Incorporated sought to determine whether it would yield better crops and profits for farmers. In a study incorporating five universities, researchers compared the quality of cotton when harvesting:

 

  • 1) Two times per week after the first boll opens
  • 2) Once by hand at the end of the season
  • 3) Once with a machine at the end of the season.

 

Results indicate that multiple harvesting improved yield and both fiber and seed quality. Multiple harvesting also created greater uniformity in cotton quality and grade, which could provide farmers with premiums and new markets. Once robotic harvesters reach their full potential, multiple harvesting might become a new agricultural benchmark.

  

This Cotton Incorporated-sponsored research was assembled from studies at Clemson, Kansas State, Texas A&M, University of Georgia, University of Tennessee, NC State and University of North Texas. Here we provide an overview of the changes to cotton production that advances in robotics might drive with multiple field studies conducted over two years. Download the white paper to learn more about 3 key ways robotics might improve yields and farmer profits in the years to come.

 

The work we do is possible because of collaborations with researchers like these and partnerships with people all throughout the value chain. Ready to commit to sustainably produced cotton? Become a Cotton LEADS partner today. Interested in doing even more? Contact us for ideas to get the most out of sustainable cotton and your partnership with Cotton LEADS.

More news