Horticulture

Tweet

Data Fusion Creates More Accurate Insights

Image Source: elenabsl/Shutterstock.com

By Lucas Costa for Mouser Electronics

Published August 9, 2021

Introduction

Now more than ever, the agriculture industry needs to do more with less. According to the World Bank, farmers and ranchers make up just 1.3 percent of employed people in the United States, meaning that just 4 million farmers feed today’s 335 million people. Farming alone contributes roughly $135 billion (USD) to the U.S. economy each year. In 2018, the agriculture sector employed around 1 billion people worldwide, accounting for 27 percent of the global workforce. This is a decrease from 44 percent in 1991. The numbers are staggering. Farmers face many challenges, with the weather, pests, demands for fewer chemicals, and more that create agriculture management headaches.

Recently, farmers have employed various technical solutions, such as using drones and remote sensors to provide insights about field and soil conditions—information that farmers use to determine watering, spraying, and seeding schedules. Part of a movement called precision farming, these technologies have provided insights that were previously unavailable or that required considerable time and effort to glean. Combining various sensor types, however, creates additional challenges. In the following, we’ll look at sensor fusion and intelligent farming systems that stand out in solving problems and helping farmers meet growing demands.

The Promise of Precision Agriculture

Imagine a farm consisting of hundreds or thousands of acres, with dozens of fields planted with hundreds of thousands of individual plants. Optimizing crops to yield the best results requires continuous assessments of—and quick response to—field conditions, moisture levels, soil chemicals, plant size and health, diseases, pests, and much more. Done manually, sampling and analysis are typically performed at the field level, which does not account for variances across a single zone or from plant to plant. Additionally, this process is time-consuming and labor-intensive, and results might not necessarily apply to all rows or individual plants within a parcel or be delivered in time to save crops. With missing or imprecise data, the manual process likely does not accurately portray conditions over time.

Advances in sensors, satellite imagery, robotics, and big-data analytics have given rise to precision agriculture, a management concept that focuses on observing, measuring, and responding to field variables in real-time and providing an accurate picture of conditions over time. When fully enabled, the approach uses technology to:

• Monitor conditions
• Assess conditions and diagnose problems
• Decide which treatments or tasks are needed to optimize crops
• Implement treatments and tasks

Although grouping crops by field region is helpful, the next challenge in precision agriculture is achieving a much higher granularity—down to the plant level. The objective is to identify the correct treatment requirements at the plant level instead of for the larger area to improve harvest and minimize waste.

Precision Agriculture Challenges

Although precision agriculture helps farmers optimize crops, processing large amounts of data from multiple collection sources creates its own challenges. To understand the needs of each field and crop, farmers need access to both large- and small-scale insights.

Large-scale data refers to influences that a particular region experiences from environmental aspects such as solar radiation, temperature, precipitation, and soil composition. This data is crucial for farmers to manage the plantation because it is an indicator of fertilizer, pesticide, and water requirements. The disadvantage of relying on this data is its collection, which involves inaccuracies from variations between individual plants. Sensors and machines capable of recognizing the individual plants achieve precision analyses, mainly in real-time, to adjust their tasks correctly.

Real-time data acquisition comes from high-precision sensors coupled with smart machines in the field. Multispectral cameras observe leaf coloration in spectra that humans cannot see. Artificial intelligence (AI) systems capable of classifying plants by age, stage of growth, and health can use images from these cameras to identify and count fruits and flowers, generating essential data about the plant that is not currently collected because doing so requires intensive labor and cost. Light detection and ranging (LiDAR) sensors capable of collecting spatial data and providing real-time measurements of crop volume and density help farmers make individualized decisions based on the current status of each plant.

The challenge of precision farming is understanding how different parameters and collected data affect the plant and implementing systems capable of processing this data in real time for field applications. Novel sensors and systems are capable of such complex analyses quickly, but their implementation requires extensive knowledge in multidisciplinary areas, with a lot of research and experimentation.

Data Fusion Solves Challenges

Given the complexity of the data collected, it’s not surprising that processing data in a way that provides the right insights poses significant challenges. Data fusion has begun to help solve these problems. The concept behind data fusion is integrating multiple sources of data to produce more accurate and helpful information than any individual data source could provide. This effect can be seen in agriculture with a soil analysis map and tree fruit count, which do not fully represent the crop condition alone but help farmers better understand plant needs, such as the correct amount of each nutrient for the highest yield.

It is possible to better understand each plant in the field by combining global positioning systems (GPS), soil maps, drone imagery, ground images, and LiDAR. Multiple systems, each equipped with sensors, are required to obtain such a large amount of data.

Sensor Fusion

Sensor fusion is the process of merging data derived from multiple sources to provide information that is more accurate than would be feasible if the sources were used separately. Machines capable of sensor fusion can use multiple sensors and a microcontroller equipped with AI and smart algorithms capable of processing all the sensor’s data, usually in real time, to provide a better output by using all readings or understanding which sensors to prioritize for each task.

In a field during a pesticide application, such a smart machine can use a camera’s images to identify the level of a disease infection per tree correctly, but it is limited by spatial understanding. In this sense, LiDAR helps describe the dimensions of the crop to adjust the volume of chemicals being applied correctly.

Previously, data collection in agriculture was so labor-intensive and expensive that farmers would have to collect their data per block or region. Sensor fusion provides an accurate assessment of the crop at the plant level, a level of precision never seen before in agriculture.

An example of a sensor fusion system currently being applied in agriculture is the SmartSense system, developed by Agriculture Intelligence of Florida. The system contains cameras, LiDAR, and GPS sensors to collect citrus crop data in real-time, such as tree height, canopy density, health status, and fruit count. This system can be attached to any ground vehicles that drive in the field, such as tractors. If connected to a citrus sprayer or spreader that typically runs on the areas multiple times a month, the data collected can be used to control the machine, providing a variable-rate methodology that collects close-up data while also adjusting the application of chemicals in real-time.

System of Systems

A system of systems, such as data fusion, is a collection of dedicated systems that pool their capabilities to produce a more sophisticated system with greater functionality and performance than the sum of its parts. Instead of merging data from multiple sensors, a system of systems uses the outputs and analyses of multiple machines to provide large-scale insight with high precision.

The need for such a complex system is easily visible in data collection for agriculture, which often requires multiple systems, such as weather stations for temperature and precipitation, satellite and drone imagery for field scouting, and water quality sensors for irrigation. Although all this information is helpful for each system to complete its task, it can also help improve farm management.

Crop management depends on fast, accurate detection of problems as early as possible. Different analyses can lead to multiple solutions to a problem, but having access to different angles can help identify the problem correctly. For example, with a soil analysis, a farmer might detect that a block has the correct amount of nutrients in the soil chemistry and should thus be healthy. The smart sprayer detects a small group of unhealthy trees during a regular application, indicating a disease. A simultaneous drone flight with a thermal camera can further the investigation, identifying the problem as lack of water from a blockage in the irrigation system.

An example of a system of systems is Agriculture Intelligence Agroview. Farmers can upload images collected from farms using a drone with spectral cameras, soil analysis maps, past harvest information, and much more to this cloud-based software. The platform generates an aerial map of the field and provides information about crop count, size, health, yield prediction, and even fertilizer requirements for individual plants.

Crop health analysis can help farmers adjust their use of pesticides, while fertility analysis identifies spots that require more or fewer fertilizer applications. Farmers can use these outputs as application maps. Smart variable-rate machines use these data outputs to identify each area’s need to adapt its application rate. This process can significantly improve cost savings, reduce pesticide use, and generate a higher-quality harvest.

In a study on smart machinery in agriculture conducted at the University of Florida, the citrus tree sprayer equipped with the SmartSense system reduced fertilizer use up to 30 percent in citrus groves in Florida With similar results in cost reduction, the SmartSense also presents the novelty of assessing health and predicting harvest yield reliably and accurately.

Conclusion

Smart, autonomous field machinery capable of variable-rate applications provides a solution to an increasing need to reduce waste and chemical use on crops. Data fusion techniques enable these systems to generate a reliable and accurate crop assessment in different fields and conditions. With limited resources, new technologies are being applied to minimize resources while continuously meeting harvest demands.

With a continuous increase in demand for larger and better harvests, while simultaneously lowering chemical use in the environment, new technologies aim to help farmers manage their crops. Precision agriculture techniques can help farmers meet demands with fewer resources by combining the best in data analysis for field applications with the most effective data-management practices to maximize yields and minimize spending.

Lucas has a Masters in Machine Learning and Robotics, and is an Artificial Intelligence developer, researcher, and practitioner as well as an Open Source Software developer, with experience in machine vision and automation for smart machinery in Precision Agriculture in his work as a researcher at the University of Florida and developer at his co-founded company Agriculture Intelligence. He has authored many papers as well as cloud software (Agroview) and data fusion systems.