cyberquantic logo header
EN-language img
FR-language img
breadcrumbs icon
Agriculture

Crop health Analysis

Real-time segmentation and prediction of plant growth dynamics using low- power embedded systems equipped with AI

Real-time segmentation and prediction of plant growth dynamics using low- power embedded systems equipped with AI

For:
Agriculture, ecology management, sanitary services
Goal:
Improve Operation Efficiency
Problem addressed
Prediction of harvest, biomass/leaf area dynamics, leaf index, parameters
describing the quality of produced food, consumption of resources from
sequences of images of plant growth (including multispectral), data from sensors
that describe environmental conditions and artificial growing system
parameters representing the state of the growing system.
Scope of use case
The project is devoted to the development of a low-power embedded system and AI algorithm for real-time plant segmentation and prediction of its growth. The proposed distributed system is aimed for use in greenhouses and remote areas, where edge-computing autonomous systems are in demand. A branch of this project also aims to develop the payload for drones for the segmentation of harmful plants in real-time.
Description
Research efforts towards low-power sensing devices with
fully-functional AI on board are still fragmented. In our
project, we present an embedded system enriched with the
AI that ensures the continuous analysis and in-situ
prediction of plant leaf growth dynamics and other
important growth parameters. The embedded solutions,
grounded on a low-power embedded sensing system with a
graphics processing unit (GPU), are able to run the neural
networks-based AI on board. Advantages of the proposed
system include portability and ease of deployment. We use a
sequence of convolutional neural network (CNN) and a
recurrent neural network (RNN) called the long-short term
memory network (LSTM) as the core of the AI in our system.
The proposed approach guarantees the system autonomous
operation for 180 d using a standard Li-ion battery. We rely
on state-of-the-art mobile graphic chips for smart analysis
and control of autonomous devices. We used 5 514 images as
a source for automated leaf area calculation and follow the
training of AI algorithms. Over one thousand records from
sensors provide additional information about environmental
conditions. All this data was used for training and testing the
recurrent neural network, convolutional neural network
algorithms, and the segmentation algorithms. Our solution
provides a RMSE close to 4 cm2 in a 3 h prediction horizon.
All this allows for high performance in-situ optimization of
plant growth dynamics and resource consumption.
Interested in the same or similar project?
Submit a request and get a free project evaluation.