AI_for_agriculture

10+ efficient AI computer vision use cases for higher yield

10+ efficient AI computer vision use cases for higher yield

There is a great need to improve crop productivity, and AI is doing it quickly and efficiently, not just in laboratories but on actual Agri holdings!

We have collected great examples of how intelligent algorithms help crop production and increase growers’ productivity during the season, from pre-planting to post-harvest. All the use cases listed require visual data (preferably from drones) and trained AI models to solve particular issues. In a nutshell, visual data provided by drones are the “eyes” of the operation, and AI and machine learning intelligence is the “brain,” together, they are changing the Agroindustry making it more productive and efficient. 

This article covers AI application for crops and vegetables such as cereals, corn, soy, wheat, sugar beet, and more, but specific use cases for orchards, indoor farms, and livestock are also feasible and beneficial, this is just a topic for another discussion.    

All is visible on the field during the season can be measured, estimated, and improved with the help of AI.

Agriculture cycle

Agriculture cycle

Crop planning

1- Waterlogging and field relief analysis. 

These details most of the time cannot be precisely defined with the human eye or satellite images. Drone images and AI can provide farmers with insights that can bring value even before the first plants appear on the field. Knowing the field’s specificity will help support numerous upcoming decisions on seeding, irrigation, inputs, etc.   

Definition of production zones

Planting  

2 – Stand count and yield assessment.

Review the precise number of plants in a specific field and make decisive actions if it’s under the norm or expected results. Evaluate seed quality (germination rate), decide whether to replant or not, assess the yield, and mark the zones of potential yield losses.   

3 – Plant density.

Since both high and low crop densities reduce yield and total revenue, it is crucial to know the situation on the field as soon as possible. Growers know there is a population for each crop that maximizes crop yield. Start strong! 

Precise plant standcount

Crop monitoring 

4 – Heading date/flowering stage detection. 

Accurate chemical thinning of some crops and fruit trees requires estimating their blooming intensity and determining the blooming peak date. With drone data, AI quantifies the size and number of flowers for timing fertilizer input.  

5 – Weed detection. 

Left uncontrolled, weeds can result in 100% yield loss. With vision processing and machine learning technology, today’s AI-driven equipment can reduce 80-90% of herbicide use. AI can distinguish and classify weeds with a high level of accuracy to support wise decisions.   

Weed detection in sunflower

6 – Disease detection. 

Up to 50% of yield loss is caused by pests and diseases every year. AI technology helps diagnose plant diseases, classify them and significantly decrease the number of chemicals applied.   

7 – Insect damage.

It is humanly impossible to distinguish insect categories and the growth period of insects without the knowledge of entomology. AI computer vision makes it possible to recognize and classify insects. It can quickly locate the information of an insect positioned in a complex background, accurately distinguish insect species with high similarity, and effectively identify the different phenotypes of the same insect species in different growth periods. This information will help reduce the number of insecticides applied and fight pests intelligently.  

8 – Nitrogen analysis.

The deficiency of this macro-nutrient directly affects crop health and yield. However, when Nitrogen inputs to the soil system exceed crop needs, excessive amounts of nitrates enter ground or surface water, thus leading to significant environmental harm. With the multispectral drone camera and intelligent, trained algorithm, staying informed and supporting decision-making on fertilizer quantity and timing is possible.   

9 – Crop monitoring and Plant stress analysis. 

Tracking healthy leaf color, plant growth, and ever-changing environmental conditions during the season requires significant time and human efforts. AI computer vision offers an effective alternative for growth stage monitoring because of its low-cost (relative to person-hours invested in manual observations) and the requirement for minimal human intervention, as a result – less costly mistakes and better yield.

10 – Specific tasks and quality control. 

As an example – missed tassel recognition during detasseling in maize seed production. Every tassel not removed by a machine leads to unwanted pollination and decreased genetic purity of maize seeds. AI can recognize missed tassels from 3 cm size before they start pollinating and help to save the whole field. 

Harvest & Postharvest  

11 – Plant size for harvest readiness.   

In practice, the crop growth stage is still primarily monitored by-eye, which is laborious and time-consuming and subjective, and error-prone. The application of AI computer vision offers a precise, cheaper alternative to manual observations. It is now easier to plan the labor needs, predict the best harvest day, and even refine revenue projections.  

Tomato ripe check

12 – Remaining plant sizes and count.   

Make sure that you do not leave the revenue behind. This second check and intelligent model will help identify the remaining plant counts and size. Last but not least – there are insights about quantifying crop residue, which help you with your further soil management strategies (like cover cropping or no-till planting).  

Final word. 

AI can provide growers with instant insights from their fields help them to identify areas that need fertilization, irrigation, or pesticide treatment. The AI technology combined with timely actions allows growers to reduce the number of fertilizers, help increase food production while minimizing resources, increase yield and profit, and make growers’ lives easier in the current labor-shortage situation. With valuable information, growers can make better decisions, recalibrate their future planting strategy, and finally grow more with confidence. 

AI use cases in Agroindustry
Missed_maize_tassel_recognition

Leveraging AI in Hybrid Seed Production

Case Study: Innovation in corn hybrid seed production

Innovation in corn hybrid seed production: Identifying missed tassels during detasseling to increase genetic purity, decrease cost and prevent lost revenue

Hybrid corn seed production problems

  • High chances of losing a field due to missed tassels in female inbred rows
  • High cost of manual detasseling due to reruns
  • The procedure must be carried out in a very limited time with the high accuracy

CHALLENGE

Machine detasseling only results in around 85% pulling vs the needed 99.7% for hybrid corn seeds

Any tassel not removed during the detasseling phase of hybrid seed corn production can result in unwanted pollination, degrading the necessary genetic purity, quality of seeds thus impacting its commercial market value.

SOLUTION

  1. Drone images uploaded to Proofminder platform. We collected data from 5 distinct fields and multiple corn varieties
  2. Field visualization. Orthomosaic of the plot is automatically created in the system
  3. High precision stand count readily established
  4. Male and female lines are distinguished based on phenology. Sowing structure, lines, distances between lines and individual plants identified without reliance on sowing or other external data. Male / female distinction allows to ‘ignore’ male tassels during missed tassels identification phase
  5. Clear report ready. Actionable insights on the palm of a hand
  6. Yield saved. With timely actions of the agronomist cost of detasseling decreased up to 1/3

DETAILS

Data Collection Approach:

  • Drone camera angle set at 60 degrees not at nadir – much larger plant area visible, training and verification is easier
  • Our algorithm calculates tassel location with 5-10cm accuracy based on drone location
  • Off-the-shelf drones (DJI Phantom, Matrice 210 v2 + Zenmuse X5S, etc.) and flight planning software (DJI Ground Station Pro) used
  • Above 100 hectares per drone per day coverage and processing possible

Identification of missed tassels:

  • Tassels categorized and predicted per size (S-XL)
  • Male tassels excluded, leveraging row identification algorithm
  • Identification is infinitely scalable using cloud resources – processing of 1 hectare on 1 node under 30 minutes
  • Identified missed tassels visualized in the Proofminder application on desktop/tablet/mobile but can be exported into geojson or other common geospatial formats

Implement precise detasseling use case on your farm

Avoid losing a field due to missed tassels

Get higher genetic purity of the hybrid maize seeds

Decrease cost of detasseling up to 1/3 with less labor hours

TAKE CONTROL OF DETASSELING WITH DATA AND AI

Book a demo to implement precise detasseling use case on your farm!
Ragweed_detection_with_AI

Case Study: Ragweed Identification​

Case Study: Ragweed Identification​

Ragweed (Ambrosia) identification based on phenology using high resolution RGB drone aerial photos

CHALLENGE

Ragweed pollen is a common allergen causing yearly suffering for millions of people. A single ragweed plant can produce up to a billion pollen grains in a season, which are carried long distances by the wind. It is a major health concern at many parts of the world – also various countries have laws triggering fines if too much ragweed is found on a property.

ragweed_detection
ambrozia_detection_computer_vision

SOLUTION

Our partner turned to Proofminder to be able to recognize and identify ragweed on many thousand hectares a day based on plant phenology. 

It took just a few weeks for visual AI development and deployment on the Proofminder platform ensured scalability right away. 

Ragweed detection, calculation of infection metrics became accessible via a simple, browser-based map display to make decisive actions.

DETAILS

  1. Orthomosaic is automatically created
  2. Multiple Machine Learning algorithms evaluated leveraging Proofminder’s MLOps quick iteration capabilities. SVM and Random Forest deemed most applicable
  3. Infection metrics accurately calculated according to local law
  4. Country level deployment envisioned using Proofminder:
  • 1st pass using satellite photos to identify potentially infected areas
  • 2nd pass uses drone photos for detailed analysis and ragweed identification
  • 3rd pass can focus on elimination – e.g. spraying
ragweed_detection