# AI Applications in Agriculture and Horticulture | llmnet.nl

[Skip to content](#lm-inhoud)Network/[NL](/en/ai-tools-voor-landbouw-en-teelt)EN[Hubhub.llmnet.nlCompare models on task, language, cost and licence.](https://hub.llmnet.nl/en/)[Communitycommunity.llmnet.nlPrompt techniques, patterns and system prompts.](https://community.llmnet.nl/en/)[APIapi.llmnet.nlLLMs in production: rate limits, routing, structured output.](https://api.llmnet.nl/en/)[Consultancyconsultancy.llmnet.nlRolling out AI in an organisation, pilot to production.](https://consultancy.llmnet.nl/en/)[Newsnieuws.llmnet.nlAI developments, explained for the Netherlands.](https://nieuws.llmnet.nl/en/)[Benchmarkbenchmark.llmnet.nlMeasure AI quality yourself, on your own tasks.](https://benchmark.llmnet.nl/en/)[Careersvacatures.llmnet.nlAI roles, salaries and career paths in the Netherlands.](https://vacatures.llmnet.nl/en/)[Learnleren.llmnet.nlAI concepts in plain language, beginner to builder.](https://leren.llmnet.nl/en/)[Guidegids.llmnet.nlRun AI privately on your own Mac, PC, NAS or home server.](https://gids.llmnet.nl/en/)[Directorydirectory.llmnet.nlMapping the AI ecosystem: tools, models, companies.](https://directory.llmnet.nl/en/)[Radarradar.llmnet.nlSignals from X, research and communities for indie developers.](https://radar.llmnet.nl/en/)[Appsapps.llmnet.nlReviews of AI apps and open-source repos, with tips for builders.](https://apps.llmnet.nl/en/)[llmnet.nl — main site](https://llmnet.nl/en/)[](https://x.com/intent/post?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt&text=AI%20Applications%20in%20Agriculture%20and%20Horticulture)[](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt)[](https://www.reddit.com/submit?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt&title=AI%20Applications%20in%20Agriculture%20and%20Horticulture)[](#)[](https://x.com/intent/post?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt&text=AI%20Applications%20in%20Agriculture%20and%20Horticulture)[](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt)[](https://www.reddit.com/submit?url=https%3A%2F%2Fdirectory.llmnet.nl%2Fen%2Fai-tools-voor-landbouw-en-teelt&title=AI%20Applications%20in%20Agriculture%20and%20Horticulture)[](#)

# AI applications in agriculture and horticulture

By Ivo Donker — compiled with AI support (Claude & Gemini) · Last updated: 6 August 2026

This overview covers the technological architecture, data flows, and operational trade-offs involved in deploying artificial intelligence in open-field farming, greenhouse horticulture, and livestock farming. The emphasis is on data quality, edge hardware, and contractual aspects.

## The physical reality of agricultural AI

The application of artificial intelligence in the agricultural sector differs fundamentally from classic IT environments and office applications. Where information systems in the financial sector or business services primarily work with structured tables and text, the input in agriculture and horticulture consists mainly of multispectral images, video streams, and continuous sensor measurements.

In addition, the physical environment places high demands on the hardware and the underlying algorithms. Equipment in the field or in the barn has to deal with dust, moisture, vibrations from agricultural machinery, temperature fluctuations, and changing light conditions. A computer vision model that functions properly under constant laboratory lighting can become unreliable when bright sunlight, cloud shadows, or splashing mud affect the camera images.

The challenge for developers of agricultural AI systems is therefore not only choosing the right network architecture, but also guaranteeing robustness under widely varying conditions outside the data center.

## Application areas in open-field farming and arable farming

In open-field farming and arable farming, artificial intelligence and machine learning are used to carry out observations that used to be done manually and on a sample basis, at a larger scale and with a higher frequency. This happens through several technical approaches.

### Crop monitoring from the air and the ground

For monitoring crop development, systems use camera images from satellites, unmanned aerial vehicles (drones), and image sensors mounted on field carts or implements. Besides standard RGB images, multispectral and hyperspectral cameras play a central role. These sensors capture wavelengths outside the spectrum visible to humans, such as near-infrared.

Based on these reflectance values, algorithms calculate vegetation indices. Deviations in these indices indicate differences in biomass, moisture content, or nitrogen uptake. By analyzing time series of these images, patterns in germination, growth rate, and ripening can be mapped across the entire field.

### Disease and pest recognition

Visual recognition models, mainly based on convolutional neural networks (CNNs) and modern vision transformers, are trained on large collections of leaf images. These models can identify symptoms of specific fungi, bacterial infections, or insect feeding damage.

The goal of these models is to locate infection hotspots before visual symptoms become observable to the naked eye on a large scale. The technological challenge lies in the variation of symptoms per crop variety and the similarity between nutrient deficiencies and disease symptoms on the leaf.

### Weed recognition and site-specific treatment

In weed management, high-resolution computer vision equipment makes it possible to distinguish between the cultivated plant and unwanted vegetation. Spray booms or mechanical weeding implements equipped with local processing units capture the soil surface row by row.

The software performs segmentation at the millisecond level, identifying each detected green object as crop or weed. The dosing system or mechanical gripper then receives a signal to act exclusively on the exact coordinates of the weed. This requires tight integration between the visual processing pipeline and the mechanical actuators.

### Yield prediction

Yield models combine historical cultivation data with current measurements. The input consists of a combination of soil maps, historical rainfall and temperature data, satellite indices, and information about the specific crop variety.

Machine learning models look for patterns between the historical environmental variables and the eventual harvest results. The predictions help with planning logistics processes, storage capacity, and processing schedules further down the chain.

## Greenhouse horticulture as a controlled environment

Greenhouse horticulture occupies a distinct position within the agricultural AI landscape. Unlike open-field cultivation, a greenhouse is a largely closed and controllable ecosystem. Environmental factors such as temperature, relative humidity, CO2 concentration, irrigation, and artificial lighting are continuously measured and adjusted.

Greenhouses have extensive networks of sensors that generate high-frequency time series. This allows AI models to work with feedback loops that are impossible in open-field farming. Where arable farming primarily relies on passive observation, AI in greenhouse horticulture focuses on active, continuous control.

Algorithms for climate and energy management analyze the interplay between outside conditions (such as solar radiation and wind), the dynamics inside the greenhouse, and the plant's physiological response behavior. Based on this, the models calculate the optimal settings for ventilation windows, screens, supplemental heating, and CO2 dosing.

Because all input variables are more controllable, control models in greenhouse horticulture achieve a higher degree of accuracy. A precondition, however, remains that the sensors are accurately calibrated; a deviation in a temperature or moisture sensor can cause the algorithm to carry out incorrect control actions.

## Animal health and behavior recognition in livestock farming

Within livestock farming, behavioral analysis forms its own application domain. The focus here is not on plant physiology but on the physical condition, welfare, and health of individual animals or herds.

Data collection takes place through various sensor types:

 
- Wearable sensors: Accelerometers in collars, ear tags, or leg bands record the movement patterns of animals. Algorithms translate this raw acceleration data into specific behaviors, such as rumination, resting, eating, and walking.
 
- Optical camera systems: Ceiling cameras above barns or at milking and feeding stations analyze the posture and gait of the animals. Motion analysis models can spot lameness or altered group dynamics at an early stage.
 
- Acoustic monitoring: Sound sensors in barns record acoustic signals. Specific algorithms filter out background noise and identify abnormal pattern sounds, such as coughing in young animals.
 
- Environmental and production sensors: Continuous measurements of milk quality, weight progression at feeding stations, and drinking water intake are linked to the behavioral data.

By flagging deviations from an individual's baseline behavior, these systems help the livestock farmer identify at an early stage which animals need extra care or a medical check.

## Categorizing the software and AI offering

The range of AI functionality within the agricultural market can be divided into four main structures, each with its own characteristics in terms of integration and management.

 
 
 Category | 
 Description | 
 Characteristic integration | 
 

 
 
 
 Built-in machine functions | 
 AI functionality that is directly integrated into the software of tractors, spray booms, or climate computers. | 
 Processes data locally on the device via closed, proprietary hardware. | 
 

 
 Integrated data platforms | 
 Software that brings together data from multiple sources (sensors, satellites, machines) in one place. | 
 Uses APIs and cloud infrastructure to combine heterogeneous data. | 
 

 
 Subscription-based advisory services | 
 External platforms that analyze measurement data and periodically produce action or task maps. | 
 Data is uploaded to a cloud environment; results are downloaded as a map file. | 
 

 
 Open models and custom development | 
 Deployment of freely accessible neural networks and algorithms for company-specific questions. | 
 Requires in-house IT capacity and integration with internal data sources. | 
 

 

When considering custom development or in-house pipelines, use is regularly made of [open-source models](https://directory.llmnet.nl/en/open-source-modellen) to perform specific vision or analysis tasks without depending on proprietary vendor platforms.

## The problem of model generalization and data overfitting

A structural challenge in applying machine learning in agricultural practice is the limited generalizability of models. An algorithm trained by a vendor on test fields in a specific region with a particular soil type does not automatically perform well on a different field with a different soil temperature, moisture regime, or crop variety.

This problem arises because models are sensitive to variables that were not present in the training dataset. Examples of disruptive factors include:

 
- Differences in soil color and organic matter content, which affect spectral reflectance.
 
- Variations in leaf geometry and color between different crop varieties.
 
- Local microclimates that cause deviating growth patterns.
 
- Shading from windbreaks or nearby buildings.

When an AI model is confronted with these unseen conditions, accuracy can decrease significantly. It is therefore necessary to validate systems on your own farm data beforehand. The importance of sound data collection and validating input data connects directly with the principles of [data quality for AI systems](https://consultancy.llmnet.nl/en/datakwaliteit-voor-ai), where the representativeness of training data is decisive for the end result.

## Connectivity problems and the need for edge AI

Many agricultural fields are located in areas where wireless network connections (such as 4G or 5G) have limited coverage or low bandwidth. This forms a direct obstacle for AI systems that depend on continuous data transfer to the cloud.

Sending high-resolution video footage from a fast-moving spray boom to an external cloud server to have a decision made there about weed dosing causes excessive latency. Moreover, processing stops if the network connection drops.

To solve this problem, processing shifts to the device itself. This principle is known as edge computing. The AI models run on specialized, energy-efficient chips installed directly on the agricultural machine or the field robot. The camera images are processed locally and converted within milliseconds into a control signal for the field sprayer or gripper.

The development of robust hardware that can withstand vibration and heat, combined with efficient algorithms, is an important part of the current [developments in robotics and AI](https://nieuws.llmnet.nl/en/robotica-en-ai) within the agricultural sector.

## Data ownership and contractual arrangements

With the increasing use of smart machines and sensors, the volume of detailed cultivation and business information is growing. This brings with it a legal and strategic discussion about data ownership and usage rights.

Sensor measurements, yield maps, and application data contain competitively sensitive information about a farm's productivity and soil quality. Vendors of AI platforms often use the collected customer data to further train and improve their central models.

In practice, the agreements on this rarely appear in the user manual, but are laid down in the terms and conditions and license agreements. Important contractual points of attention are:

 
- Ownership of raw data: Does the farmer remain the legal and operational owner of the raw measurement values generated on their own farm?
 
- Right to reuse by vendor: Does the vendor have the right to use the data, anonymized or aggregated, to develop new commercial services?
 
- Data portability: Can the user export the collected historical data in a standard format when switching to a different platform?
 
- Access by third parties: Under what conditions is data shared with suppliers, buyers, or government authorities?

Carefully checking these terms prevents a farm from becoming dependent on a single vendor or unintentionally transferring valuable operational data.

## The role of large language models in agricultural practice

Although image processing and sensor analysis form the largest part of agricultural AI applications in the field and in the greenhouse, large language models (LLMs) fulfill a specific role in knowledge access and business operations.

The agricultural sector deals with a large volume of documentation, including cultivation guidelines, safety data sheets for crop protection products, technical manuals for machinery, and a complex system of national and European laws and regulations. Language models are used as search and query systems on these document collections.

By using [multimodal models](https://hub.llmnet.nl/en/multimodale-modellen-overzicht) employees can combine textual documents with photos of label instructions or damage patterns. A user can ask a specific question about the permitted dosage of a product under certain conditions, after which the system looks up and summarizes the relevant article from the official documentation.

It is essential here that the language models are configured to answer exclusively based on verified sources, to prevent the model from generating incorrect or outdated regulations.

## A practical starting point for small and medium-sized businesses

For small to medium-sized agricultural businesses, purchasing highly advanced autonomous implements or rolling out a comprehensive network of wireless sensors is often too large an investment. The step toward artificial intelligence does not, however, have to start with hardware in the field.

In practice, the first concrete gains are achieved by digitizing and structuring administrative and operational processes. Automatic recording of cultivation activities, hours accounting, certification documentation, and inventory management forms the foundation on which later AI applications build.

By initially using general-purpose [data analysis tools](https://directory.llmnet.nl/en/ai-tools-data-analyse) a business can start analyzing existing data flows, such as historical energy estimates, invoicing data, and yield figures per field. Only once this administrative data flow is in order does it become effective to consider investments in physical sensors and field algorithms.

## Further reading

 
- [AI tools for the supply chain](https://directory.llmnet.nl/en/ai-tools-voor-supply-chain)
 
- [AI Tools for Data Analysis](https://directory.llmnet.nl/en/ai-tools-data-analyse)
 
- [Overview of open-source models](https://directory.llmnet.nl/en/open-source-modellen)
 
- [Overview of multimodal models](https://hub.llmnet.nl/en/multimodale-modellen-overzicht)
 
- [Data quality for AI systems](https://consultancy.llmnet.nl/en/datakwaliteit-voor-ai)
 
- [Developments in robotics and AI](https://nieuws.llmnet.nl/en/robotica-en-ai)

llmnet.nl - overview of the AI ecosystem