Palm tree icon

End-to-end Solution for Date Palm Growth

Together with a large international trading group, DFG has developed a one-of-a-kind proprietary technology for remote analysis of soil & plants on open date palms plantations in Africa and the Middle East.

End-to-end Solution for Date Palm Growth Image End-to-end Solution for Date Palm Growth Image End-to-end Solution for Date Palm Growth Image End-to-end Solution for Date Palm Growth Image



About this project

Our customer is a company, designing cutting-edge technologies for growing plants in the Middle East region. The regional cultivation specifics are very demanding towards the water and forage culture: the resource management of expensive water and fertilizers must be used carefully to avoid severe losses. According to the World Economic Forum’s data, seasonal consumption in the Middle East will grow due to global climate changes.

Our portfolio has many successfully implemented precision agriculture projects. That’s why we already knew how to approach an issue when we received a request to develop a remote analysis technology to study soil properties at open long-distance plantations. Our plans underpinned the earlier devised IoT-based system decisions and the AI-driven system of plant growth conditions.

The core challenge of the project dealt with its massive scale — the methods and technologies working for smaller lands would not be appropriate for the soil analysis on the areas from 100 ha. Why so? It’s more economically viable to use a few sensors on smaller open areas and visually inspect them from time to time.

In the beginning, we realized that this project couldn’t be limited by the in-door development. More than three months of field tests resulted in the successful launch of the cutting-edge proprietary technology further implemented by our partner.

Project Goals

Project Challenges

1. Lack of a ready-to-go and obvious solution for that land scale

It’s usually not sensible to set up separate sensors throughout big fields, which is expensive and doesn’t provide enough data for building a whole picture. It could seem that drones or satellite imagery could assist us with our task. However, during the design and discovery phase, these raw data extraction methods demonstrated significant limitations. That’s why we needed our technology that was going to be developed.

2. Disparity of data sources is huge

The data sources are represented by photos and videos taken by drones, images from satellites, data from sensors, and statistical information from Internet sources (for example, weather and statistics). All this data is received in different formats and at different speeds. It takes a lot of effort for a data engineer to match (synch)  this data both in time and in other ways.

Data from different sources needed additional processing or adjustment. For example, an optical channel required the use of neural networks such as RestNet and DenseNet to improve the quality of image classification in different ranges of the spectrum. To analyze radio signals, we needed to set up special noise-removing filters.

Fast Fourier Transforms (FFT), or Discrete Wavelet Transform (DWT), are the classic solutions for radio signal processing. They did not bring the desired result. Therefore, we resorted to other methods of obtaining data — methods of probabilistic programming and fuzzy logic. This allowed us to fix the weak predictability of signal parameters.

To come up with more accurate results, we consulted with experts and shaped the data map, taking into account their assessment and models suggested by neural network ensembles. This approach makes it possible to take into account the local characteristics of growing crops. Sometimes the opinion of expert agronomists plays a critical role in improving the efficiency of analytical modules.

Design and Discovery Phase

Agro scientists from different countries were invited to join our research efforts throughout the whole development. Their domain knowledge of date palm cultivation highly contributed to the project.

To assess the quality and quantity of vegetation in the selected area of ​​the field, the NDVI index is calculated. So far, it is the most popular method of remote vegetation analysis. The index is derived based on satellite and aerial photography data.

What is NDVI?

NDVI (Normalized Difference Vegetation Index) — is the index (with a range from - 1 to 1), showing the quality and quantity of vegetation in a certain area of the field.The index calculation principle implies that a healthy plant contains more chlorophyll than a damaged plant. It also has a higher-quality structure. Therefore, it actively consumes the red color waves and reflects them in the near-infrared region.

Alternative Solutions

While doing our research, we concentrated our main effort on finding and collecting information about possible technologies and their applications in different conditions. We’ve come up with the following requirements for a working solution:

Let’s briefly go through the additional methods of remote analysis that were not chosen due to several reasons. Below we’ll explain which exactly.

  • 1. Satellite images

    As we know, many platforms offer satellite images and image processing services. However, weather fluctuations heavily distort the accuracy of information. Also, the frequency of updates is far from ideal and we remember that the relevance of data was one of the project goals.

    In critical situations, when the weather or pests cause trouble to farmers, it is crucial to receive up-to-date information at least once a day.

  • 2. Drones

    The alternative option is an unmanned drone equipped with a multispectral camera. This alternative has downsides, too. A camera-equipped drone is a challenging item to handle. A drone takes special skills and knowledge to control, especially in difficult weather conditions. For this very reason, the adaptation of drones in agriculture is slow.

  • 3. Sensors

    Sensors are an inexpensive and efficient solution for small plantations. Installation of a sensor network to a field exceeding 50 hectares already seems hard, not to mention wheat, rice, or date palms fields over 5 - 10 thousand hectares. The main difficulties of using sensors are communication with the server, maintenance, and calibration. That is a very consuming process. Imagine if we need to run hundreds and thousands of devices with all of them needing calibration of the parameters a few times a year.

The resulting solution

As it appears, sensors, drones with cameras, and satellite imagery services wouldn’t be able to satisfy us as self-sufficient data sources.

Our working hypothesis suggested that the concurrent use of several data sources will supply us with more accurate data.

Therefore, we chose the following solution: the combination of several underground sensors, an autonomous drone equipped with a micro radar, a video camera, and the AI-enabled software developed by us. Our engineers modernized each component of this stack.

Our choice — micro radar technologies

Fresnel lens

Objects would differently reflect radio signals due to their different compositions. It works for both optical and the radio spectrum.

Our hypothesis suggested that the mineral composition and soil moisture of damaged and healthy plants will reflect radio signals differently. That is obvious with the frequency range from 2 to 60 GHz.

We combined information from radio and optical channels. After that, we tried to extract the data we needed with the help of Machine Learning algorithms.

At the experimental stage, we tested micro radars from a dozen of manufacturers and finally decided on the two devices. Various factors defined our choice: stability of operation, quality of documentation & support, ease of integration, etc.

Moreover, we devised a special microwave antenna to amplify the radio signal and increase the resolution. We used a radio Fresnel lens because it is small and easy to build.

The choice and action principle of equipment


The platform that we used was a flying-wing-type aerial drone. We chose it due to the possibility to leverage flight time, energy consumption, and equipment cost.

The data received from the drone is not broadcasted online but accumulated in the internal memory. Once the drone has landed, the data from the memory card is transferred to the local computer and processed.

To simplify drone use, we automated 80% of the actions. An agro scientist only has to select a route, press the start button, and launch the drone with their hands. The approach helps to embrace a lot more potential customers and simplifies the daily use of the device.

The drone flies to collect data and performs on-site analysis automatically. This occurs without the man’s participation. The flight can last for several hours. This duration is enough to collect information from massive areas.

At this stage, the main difficulty was to ensure that the drone enjoys a stable flight at 5 - 10 m above the tops of the plants. This height allows increasing the resolution of sensors and keeping the backlight/activation unit stable based on scanning lasers.

When a drone flies low, it is subject to high turbulence and, therefore, it must rely on high-quality autopilot. To improve flight stability, we made several constructive changes to the drone design.

AI-enabled modules

The next technological component of our equipment is AI modules used for analyzing the source information. We needed to collect more than 800 GB of data to train neural networks.

Our main goal was to improve the accuracy of identifying low moisture, low soil organic matter, and potential pest infestation. We could obtain stable results due to the combination of data from radio-locating and optical channels and underground sensors.

For example, it only took us four ground-based sensors and one drone to cover 500 hectares of field. Our team built a daily NDVI-based map showing the distribution of moisture, soil mineralogical composition, and possible infection with diseases. Afterward, we verified the information for reliability and accuracy using stationary-certified instruments.

Our experiments utilized modern advanced equipment and lasted more than three months. The main goal of tests was to safely find out the parameters in question.

To this end, we worked to determine the operation mode, modulation, and frequency of signals emitted by mirco-radars. This way, we managed to retrieve sustainable data about soil and plants with the account of the growth phase of plants.

As for the optical range, we have found a combined technical solution as an alternative to expensive multispectral cameras. We applied a combination of various cameras and special filters. As a result, the entire solution cost was reduced by 10 times.

Project Results

The main result of the project is the development of one-of-a-kind proprietary technology used for remote soil and plant analysis on a large land area. The established approach can be applied to analyze the condition of farms, cultivating date palms, rice, vineyards, vegetables, and fruit.

The cost of the technology solution was reduced by ten times due to a combination of cameras and special filters. This effect applies to equipment installation, maintenance, and work implementation.

Additionally, utilization of this technology can significantly reduce the cost of fertilizers by 25 - 30%, cut pesticide expenditure, and optimize water consumption on irrigation. The saving effect was achieved thanks to a more advanced technology used to detect soil and date palms’ needs, as well as obtaining more precise data from the combination of sources.

The saving effect was achieved thanks to a more advanced technology used to detect soil and date palms’ needs, as well as obtaining more precise data from the combination of sources.