Page 27 - AI Ready – Analysis Towards a Standardized Readiness Framework

AI Ready – Analysis Towards a Standardized Readiness Framework

and air quality and soil fertility, are also considered. Technologies like LoRa, LoRa-WAN, RFM69,
Bluetooth, and narrow-band IoT facilitate robust communication and low-power operations,
while AI, ML, and emerging technologies like 6G enhance data analysis and decision-making
capabilities. Ensuring interoperability between different sensors and communication systems
is crucial, as is incorporating farmers' experiences and practices to refine and adopt these
technologies effectively.

4�3�8 Intelligent UAV-Assisted Plant Disease Detection in Rock Melon
Greenhouses

This use case [60] addresses the problem of plant disease detection and optimal resource
allocation in melon greenhouses through a UAV-assisted model. The data used in this use case
consists of plant leaf images, collected by unmanned aerial vehicles equipped with cameras.
Data collection involves drones capturing images of leaves from different angles and heights,
enhancing detail, and improving model robustness through data augmentation. The images
are then pre-processed, labelled, and categorized using the YOLOv9 model according to
various disease categories.

Model training is based on the processed data, with ongoing feedback from farmers to test
and evaluate the model's performance. The real-world deployment in a Melon Greenhouse
ensures the collection of high-quality data and effective disease classification.

4�3�9 Digital Twins for AI-based xApps in Open RAN for Smart Agriculture
in 5G

This use case [61] using digital twins for validation of xApps and encapsulates vertical applications
in the form of xApps in Open RAN in sandbox in the context of 5G and 6G. The data used in
this use case is publicly available. Architectures compliant with ITU-T-Y.3172 [63], ITU-T Y.3179
[62], and ITU-T Y.3181 [64] frameworks are used to route data to the digital twin for validation.

Initially, a training pipeline is set up involving model selection, hosting the model in an xApp, and
validation within a sandbox using digital twins and simulators. Subsequently, configuration and
deployment in the digital twin occur, involving intent-based selection of xApps, data models,
model selection, and sandbox configuration. Once the verification is completed in the sandbox,
an inference pipeline is established where data is collected and sent to the Distributed Unit
(DU), and inference is performed within an Open RAN xApp that hosts the real-time model.

4�3�10 AI-enabled Soil Analysis and Weather Station for Local Farmers in
Ghana

This use case [66] analyses inaccurate weather and soil condition predictions for local farmlands
in Ghana. National meteorological data is typically not specific to local areas and hence usually
not helpful in reducing economic losses. The use case supplements global with real-time data
collection using AI-enabled weather stations and soil analysis sensors. This data provided by
CESM [65] is used to train a tinyML model [21] in the weather station challenge [67] 2024.
The sound of rain and wind collected by microphones could be analysed thus enabling the
prediction of rain intensity, precipitation, wind speed, and wind direction.

20