Agriculture and food

Farmers make use of a variety of data sources to analyse and evaluate the performance of their farming activities and inform their decision-making. Machine learning techniques can be used to collate and analyse large quantities of agricultural data, allowing farmers to better-understand, predict and manage their resources to maximise their farm production and profitability.

    Traditional agricultural data sources include:

    • Operational and business data (including income, ordering, purchasing and operational costs)
    • Commodity data (pricing)
    • Observations 
    • Local knowledge.

    As agricultural sensors (e.g. physical sensors, cameras, satellite imagery and drones) proliferate and the volume of agricultural data available to farmers increases, farmers need a way to better gather and make-sense of this data to extract the information that they need to support their decision-making. This use of technology to support farming activities is often referred to as "precision agriculture" or "smart farming".

      agriculture data to decisions

      Figure 1 - Famers collect data and make observations to inform their decision-making.

      Machine learning, with its ability to take-in and quickly analyse vast quantities of data, can be used in a variety of ways in agriculture in:

      • Sensing and monitoring - to provide accurate, real-time observations of farm conditions and activities
      • Data management and analysis - to collect and analyse data from a range of diverse data sources to provide an accurate picture of their farm status and trends
      • Farm automation - to automatically carry out farming activities.
      • Sensing and monitoring

        Sensing and monitoring refers to the use of agricultural sensors to measure the actual performance of the farm and farm processes. Traditionally, this has been done manually by a human observer(s). Farmers across the world are adopting modern technologies and equipment to provide accurate, real-time observations of farm conditions and to measure the performance of their farm processes; aiding in the prediction of yield and the early detection of disease, dehydration and pests. 

        As an example, GPS sensors have been incorporated in farm machinery since the late 1990s and have since become an essential feature of farm equipment. These GPS sensors have proven themselves invaluable as they allow farmers to determine not only where farm machinery is, but also where it has been allowing farmers to see what fields have been ploughed, sprayed, and so forth. 

        Farm machines are increasingly being equipped with sensors and cameras to capture data such as soil moisture, leaf greenness, temperature, seeding density, fertiliser and pesticide spraying rate, crop yield, fuel usage and machine performance. Automated weather stations and distributed internet-connected sensors can provide an up-to-the-minute measurement of specific parameters. Cameras, drones and satellite imagery are a relatively cheap and very accessible source of visual data on a farm’s performance. Multi-spectral cameras can provide information about crop maturity, weeds and disease.

        Farmers must analyse the data from these modern sensing technologies to extract the information that they need to inform their decision-making. Data analysis can be a very time-consuming and specialised task. Machine learning can be used to automate data analysis tasks, providing near real-time insights into farm performance.

        Computer vision and deep learning techniques can be used to train computers to look for specific physical features (e.g. field boundaries, indications of disease, the presence of weeds or the presence and ripeness of fruit) in images from cameras, drones and satellites. In this way, machine learning can be used to automate time-consuming image processing and measurement tasks. 

        Machine learning can also be used to automate time-consuming data analysis tasks.  For example:

        • Generating and collating data logs
        • Converting yield vs GPS location into yield maps
        • Creating soil sampling maps
        • Generating maps from satellite images
        • Track the location of farm vehicles
        • Calculate farm vehicle performance (e.g. ploughing efficiency)
        • Monitoring crop growth
        • Detect the presence of weeds and pests
        • Tracking the application of fertilisers and pesticides
        • Tracking the location and movement of tagged livestock.

        How machine learning can be applied is very much dependent on the specific application and the data that is available for training. AIML can work with you to find the best way to capture and interpret your data, contact us to find out how your organisation can benefit from machine learning.

      • Data management and analysis

        Farmers make a continuous series of decisions on how best to run their farms. Farmers evaluate available data and draw upon their own experience to make decisions and determine an appropriate course of action. The volume, diversity and complexity of the agricultural data is rapidly increasing and, in many cases, exceeds a human operator’s ability to recognise patterns. 

        There is a wealth of data that is available to farmers. The format of this data varies from:

        • Business data (typically structured)
        • Observations
        • Sensor logs
        • Imagery
        • GPS data
        • Historical data (unstructured).

        Machine learning can be used to collate data from a range of diverse data sources, in a variety of formats, to provide farmers with an accurate picture of their farm status. Machine learning is very effective at recognising patterns and deriving correlations. By applying machine learning to agricultural data, it is possible to extract important details quickly and efficiently, providing the farmer with an up-to-date picture of farm status and identifying trends. Machine learning can be used to:

        • Collate and interpret agricultural data to provide relevant and concise information to be considered in decision-making 
        • Acquire external data to complement direct observations
        • Recognise patterns and finding irregularities in the data, which can be used to inform the farmer of trends and providing alerts when anomalies are detected 
        • Suggest where intervention may be warranted and the best course of action. 
        • Provide alerts where farm performance deviates from the normal operating parameters
        • Plan intervention activities to correct farm processes or performance
        • Provide alerts of diseases, pests or extreme weather events 
        • Plan labour and logistics requirements
        • Undertake predictive market analysis
        • Quantify the effect of significant events (e.g. extreme weather event) on-farm processes and performance
        • Monitor the immediate and longer-term impacts of climate change on-farm performance.
        Concept image for AI in agriculture

        Figure 2 - Concept for the use of machine learning in agriculture; providing the farmer with an up-to-date picture of farm status.

      • Farm automation

        Automation is when a process or procedure is performed with minimal human assistance or intervention. A robot is a machine capable of operating autonomously and carrying out a complex series of actions. Benefits of automation include labour, energy and material savings as well as improvements in quality, accuracy, and precision.

        An autonomous software agent automatically collects and processes data, derives conclusions and provides advice. Machine learning can take in and process vast quantities of data and analyse trends and will form the core of future automated agricultural decision support technologies. These automated agricultural decision support technologies will automatically extract the necessary information to help farmers make sense of their data, driving real-time operational decisions and providing predictive insights into future outcomes.

        Physical automation may include: 

        • Control systems to monitor and modify the physical state of a system 
        • Autonomous vehicles (e.g. harvesters, sprayers and feeders)
        • Farming robots to perform specific physical tasks.

        Control systems may be as simple as a single-parameter controller (e.g. thermostat) or as complex as a large-scale control system with multiple interdependencies between the variables. Machine learning can be used to model and control the physical system.

        Autonomous vehicles are capable of sensing their environment and moving with little (partially-autonomous) or no human input (fully-autonomous). Machine learning forms the foundation for the sensing and control of autonomous vehicles. Autonomous vehicles are entering service in controlled environments where vehicles can follow predetermined, fixed routes. Specific applications in agriculture include crop planting and harvesting, weed control and livestock control and feeding.

        Agricultural robots are being developed to automate repetitive and time-consuming tasks. Robots need to be able to sense and navigate their environment and perform actions at set locations. Some of the common applications for robots in agriculture include: 

        • Planting, monitoring and caring for plants
        • Harvesting 
        • Weed control.


      AIML has worked with Bayer, Australian Grain Technologies (AGT) and Riverland Wine to develop machine learning techniques to:

      • Automatically analyse cereal crop growth
      • Predict cereal crop yield
      • Inform genomic selection for plant breeding
      • Collate and analyse data related to plant breeding, grain quality, plant genetics and trials-site data
      • Develop a status monitoring system using data from airborne and ground level cameras, discrete sensors and machine learning techniques to identify and monitor bud count, yield and canopy size.
      • Development of a digital assistant that uses machine learning algorithms to provide information on farm status and recommended actions (e.g. water, fertiliser and pesticide application, canopy management, harvest timing) based on predicted volume, quality and economic outcomes.

      How machine learning can be applied is very much dependent on the specific application and the data that is available for training. AIML can work with you to find the best way to capture and interpret your data, contact AIML to find out how.

      Agriculture and food AI in action

      The science of flavour: why food researchers are interested in machine learning

      Food and drink like you've never experienced could be around the corner with advancements in flavour and aroma prediction and formulation using machine learning!