Designing a framework for Sustainable Smart Agriculture with Mobile Robots and CNN

34 Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026
Designing a Framework for Sustainable Smart Agriculture With Mobile
Robots and CNN

Ting Zhang
University of Houston at Downtown

Oanh Phan
University of Houston at Downtown

Jiang Lu
University of Houston at Clear Lake

Advances in artificial intelligence, robotics, and computer vision enable
intelligent gardening systems that reduce manual monitoring. This study
presents a smart gardening framework integrating a mobile robot with
convolutional neural network (CNN) flower recognition. The robot captures
garden images, while the CNN classifies flower species and provides plant-
specific information for future care decisions. Using 50 training images and
50 independent testing images, the model achieved 92.0% training accuracy
and 95.0% testing accuracy. The framework connects robotic image
acquisition with automated visual recognition and provides a foundation for
future irrigation, fertilization, and autonomous garden- management
functions.

Keywords: smart devices, intelligent algorithms, robotics, CNN

INTRODUCTION

Recent advances in artificial intelligence (AI), robotics, computer
vision, and Internet of Things (IoT) technologies have created new
opportunities for automating agricultural and gardening activities.
Traditional gardening requires continuous human invol vement in plant
monitoring, watering, fertilization, disease inspection, and general
maintenance. These activities can become time- consuming when a garden
contains numerous plants with different environmental and nutritional
requirements. Intelligent sensi ng and autonomous robotic technologies
provide an alternative approach in which information about plants can be
Submitted:
July 1, 2026
Revised:
August 5, 2026
Accepted:
September 7, 2026
Published:
September 28, 2026

Citation:
Zhang, T., Phan, O., & Lu,
J. (2026). Designing a
framework for Sustainable
Smart Agriculture With
Mobile Robots and CNN.
Journal of Strategic
Innovation and
Sustainability, 21(3), 34–
41.

Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026 35
collected and analyzed automatically (Vougioukas, 2019; Yang et al., 2023; Bechar & Vigneault, 2016).
Such technologies have contributed to precision agriculture, where sensing, computation, and
automated equipment improve the efficiency of agricultural operations (Bechar & Vigneault, 2016). Smart
gardening can be regarded as a smaller -scale extension of p recision agriculture. Rather than treating an
entire garden as a uniform environment, an intelligent gardening system can obtain information about
individual plants or specific regions and use these observations to support plant-care decisions. Automated
irrigation is one important application because different plant species may have different water
requirements. Similar considerations apply to fertilizer application, environmental monitoring, and plant -
health management. Integrating sensing with automated control can therefore reduce repetitive manual
activities while allowing resources to be applied according to observed conditions (Liakos et al., 2018;
Kamilaris & Prenafeta-Boldú, 2018).
A major requirement for such a system is the ability to determine what plants are present in the
monitored area. An autonomous gardening robot may capture many images as it moves through a garden;
however, these images become useful for plant-specific operations only when the system can interpret their
visual content. Conventional image -processing techniques commonly depend on manually designed
features such as color, texture, shape, and edges. These approaches may perform adequately under
controlled conditions but can become less reliable when illumination, viewing angle, background, plant
size, or appearance changes (Wäldchen & Mäder, 2018). Gardens present additional difficulties because
leaves, stems, flowers, soil, grass, and surrounding objects may overlap within the same image.
Deep learning has significantly changed visual recognition. Convolutional neural networks (CNNs) can
learn hierarchical visual features directly from training images rather than depending entirely on manually
selected characteristics (LeCun et al., 2015; Krizhevsky et al., 2012). CNN architectures have demonstrated
strong performance in image classification and have been adapted to agricultural applications. Larger image
datasets, increased computational capability, and transfer learning have further expanded the practical use
of deep neural networks for recognizing objects under different visual conditions (He et al., 2016).
Plant identification is a fine-grained visual classification problem because different species may exhibit
similar characteristics, while plants of the same species can vary with growth stage, season, environmental
conditions, and image perspective. PlantC LEF research has supported deep -learning approaches for
identifying plants from flowers, leaves, fruits, stems, and whole -plant images (Goëau et al., 2017). Deep
neural networks incorporating taxonomic information have also been used to distinguish plants at family,
genus, and species levels (Wu et al., 2019). Computer vision has also been applied to plant -health
assessment and disease recognition (Mohanty et al., 2016; Saleem et al., 2019). Agricultural robots require
such perception systems to recognize r elevant objects before physical operations can be performed (Sa et
al., 2016).

MATERIALS AND METHODS

Design Objectives
The primary objective of the proposed smart gardening system is to integrate robotic mobility with
deep-learning-based image recognition for automated plant monitoring and identification. The system
enables a mobile robot to collect garden images, recognize flower species using a CNN model, and provide
classification results for plant -specific management. Additional objectives include reducing manual
monitoring, supporting more efficient irrigation and fertilization, and establishing a modular framework
that can incorporate environmental and soil sensors in future development. Ultimately, the system provides
a foundation for intelligent and autonomous garden maintenance (See Figure 1).

36 Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026
FIGURE 1
DESIGN OBJECTIVES

System Design and Experimental Steps
The proposed smart gardening system integrates a mobile robotic platform, an image -acquisition
module, and a CNN-based flower recognition model. The mobile robot serves as the physical platform for
navigating through the garden and collecting visual information from plants at different locations. A camera
mounted on the robotic platform captures flower images during operation. The acquired images are
transferred to the recognition module, where preprocessing standardizes the input dimensions and prepares
the images for CNN analysis.
A labeled image dataset containing multiple flower categories, including roses, irises, and other
representative garden species, is used to train and evaluate the classification model. During training, the
CNN learns discriminative visual characteristics a ssociated with each flower category, while previously
unseen images evaluate its ability to recognize individual species. During robotic operation, newly captured
images are processed by the trained network, which generates a predicted flower class and cor responding
classification information.
The identified plant category is provided to the garden-management module as an input for determining
plant-specific actions. Different flower classes can be associated with predefined irrigation and fertilization
requirements. The overall sequence consists of robotic navigation, image acquisition, preprocessing, CNN-
based feature learning and classification, plant identification, and generation of information for irrigation
or fertilizer management (See Figure 2). The modular design separates physical data collection from visual
analysis and plant -care decision making, allowing each component to be modified or expanded
independently.

Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026 37
FIGURE 2
SYSTEM DESIGN AND EXPERIMENTAL STEPS

The Hardware and Software Co-Design
The hardware platform utilized in this study is the ELEGOO mobile robot system, a compact wheel-
based robotic platform suitable for autonomous navigation, sensing, and image -acquisition experiments.
The robot employs DC motors to drive its wheels, enabling forward and backward movement as well as
directional control within the garden environment. Mobile robotic platforms provide an effective approach
for agricultural monitoring because they can collect information from multiple locations while reducing the
need for stationary sensing equipment (Bechar & Vigneault, 2016).
The robot is equipped with a camera that serves as the primary visual sensor. During operation, the
robot moves through the experimental environment and captures images of flowers from different locations

38 Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026
and viewing positions. The resulting visual data are transferred to the image -processing and classification
module. Similar combinations of mobile robotic platforms and visual sensing have been investigated for
agricultural monitoring, object detection, and automated field operations (Vougioukas, 2019).
The software component was developed in Python and employs a CNN for image -based flower
classification. CNNs are suitable for visual recognition because convolutional layers can automatically
extract hierarchical characteristics such as edges, shapes, text ures, and more complex visual patterns
(LeCun et al., 2015). The experimental dataset is divided into independent training and testing groups. A
total of 50 labeled flower images are used for model training, and an additional 50 images are reserved for
testing. The testing images are not used to update model parameters. The classification output provides the
predicted flower species and can subsequently support plant -specific irrigation, fertilization, and
maintenance decisions (Kamilaris & Prenafeta-Boldú, 2018).

RESULTS

ELEGOO Robot and Images
The ELEGOO robot successfully captured garden images using its built -in web camera under remote
human monitoring. Commands for movement, stopping, and image capture were controlled during the
experiment. Figure 3 shows the robot operating in a garden environment (See Figure 3).

FIGURE 3
ELEGOO ROBOT MOVING IN THE GARDEN

CNN Performance Results
The CNN was trained and tested using two independent sets of flower images to evaluate its ability to
recognize different species. The training dataset consisted of 50 labeled images representing selected
categories, including different flower species. Before training, the images were resized to a consistent input
dimension and normalized. During training, convolutional layers extracted visual characteristics such as
edges, shapes, colors, textures, and higher-level flower features, while network parameters were iteratively
adjusted to minimize classification error.

Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026 39
After training, a separate set of 50 flower images was used exclusively for testing. None of the testing
images were included in the training dataset. Each testing image was processed using the trained CNN, and
the predicted category was compared with its ground- truth label. This procedure provides a direct
evaluation of the recognition module before integration with the mobile robot for real -time flower
identification. Figure 4 presents model accuracy and Figure 5 presents model loss (See Figures 4 and 5).

FIGURE 4
MODEL ACCURACY OF THE CNN

FIGURE 5
MODEL LOSS OF THE CNN

Classification Results
The CNN-based flower recognition algorithm demonstrated strong classification performance. The
model achieved a training accuracy of 92.0% and a testing accuracy of 95.0%, indicating that the trained
network generalized effectively to flower images not included in training. Representative testing samples,
including daisy, dandelion, and tulip, were correctly classified, with predicted categories matching their

40 Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026
ground-truth labels. The results indicate that the network learned useful visual characteristics despite
variations in flower shape, color, orientation, and background. The slightly higher testing accuracy may
occur with a relatively small dataset and differences in sample difficulty between the two sets. Overall, the
95.0% testing accuracy demonstrates the potential of the recognition module for integration with the mobile
robotic platform (See Figure 6).

FIGURE 6
CLASSIFICATION RESULTS

CONCLUSIONS, DISCUSSIONS AND FUTURE WORK

The proposed framework demonstrates how robotic mobility and artificial intelligence can be combined
to support plant-specific garden management. Instead of relying entirely on manual observation, the robot
can collect visual information from different garden locations and use the trained CNN model to determine
the corresponding flower category. The recognition output can then serve as an input to higher-level garden-
management functions. This approach establishes a connection between plant perception and a utomated
physical operations and provides a foundation for developing more intelligent gardening systems.
Future work will focus on extending the current recognition system to directly command irrigation and
fertilization operations. After a plant is identified, its predicted species will be associated with predefined
water and fertilizer requirements. The gar den-management module will use this information to generate
control commands for irrigation pumps, valves, and fertilizer -delivery mechanisms, allowing different
plants to receive appropriate treatment. Additional sensors, including soil -moisture, temperature, and
humidity sensors, can be incorporated to combine visual recognition with real -time environmental
conditions. Future experiments will expand the flower -image dataset and evaluate the system under
different lighting conditions, viewing angles, plant growth stages, and outdoor environments. These
developments will move the framework toward a fully autonomous smart gardening system capable of plant
identification, condition monitoring, irrigation, and fertilization with reduced human intervention.

Journal of Strategic Innovation and Sustainability Vol. 21(3) 2026 41
REFERENCES

Bechar, A., & Vigneault, C. (2016). Agricultural robots for field operations: Concepts and components.
Biosystems Engineering, 149, 94–111. https://doi.org/10.1016/j.biosystemseng.2016.06.014
Ferentinos, K.P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and
Electronics in Agriculture, 145, 311–318. https://doi.org/10.1016/j.compag.2018.01.009
Goëau, H., Bonnet, P., & Joly, A. (2017). Plant identification based on noisy web data: The amazing
performance of deep learning (LifeCLEF 2017). In CLEF 2017 working notes. Dublin, Ireland.
He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 770–
778). https://doi.org/10.1109/CVPR.2016.90
Kamilaris, A., & Prenafeta-Boldú, F.X. (2018). Deep learning in agriculture: A survey. Computers and
Electronics in Agriculture, 147, 70–90. https://doi.org/10.1016/j.compag.2018.02.016
Krizhevsky, A., Sutskever, I., & Hinton, G.E. (2012). ImageNet classification with deep convolutional
neural networks. Advances in Neural Information Processing Systems, 25, 1097–1105.
LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444.
https://doi.org/10.1038/nature14539
Liakos, K.G., Busato, P., Moshou, D., Pearson, S., & Bochtis, D. (2018). Machine learning in agriculture:
A review. Sensors, 18(8), 2674. https://doi.org/10.3390/s18082674
Mohanty, S.P., Hughes, D.P., & Salathé, M. (2016). Using deep learning for image-based plant disease
detection. Frontiers in Plant Science, 7, 1419. https://doi.org/10.3389/fpls.2016.01419
Sa, I., Ge, Z., Dayoub, F., Upcroft, B., Perez, T., & McCool, C. (2016). DeepFruits: A fruit detection
system using deep neural networks. Sensors, 16(8), 1222. https://doi.org/10.3390/s16081222
Saleem, M.H., Potgieter, J., & Arif, K.M. (2019). Plant disease detection and classification by deep
learning. Plants, 8(11), 468. https://doi.org/10.3390/plants8110468
Vougioukas, S.G. (2019). Agricultural robotics. Annual Review of Control, Robotics, and Autonomous
Systems, 2, 365–392. https://doi.org/10.1146/annurev-control-053018-023617
Wäldchen, J., & Mäder, P. (2018). Plant species identification using computer vision techniques: A
systematic literature review. Archives of Computational Methods in Engineering, 25, 507–543.
https://doi.org/10.1007/s11831-016-9206-z
Wu, D., Xiang, Y., Jiang, B., & Zhang, J. (2019). Deep learning with taxonomic loss for plant
identification. Computational Intelligence and Neuroscience, 2019, 2015017.
Yang, Q., Du, X., Wang, Z., Meng, Z., Ma, Z., & Zhang, Q. (2023). A review of core agricultural robot
technologies for crop productions. Computers and Electronics in Agriculture, 206, 107701.
https://doi.org/10.1016/j.compag.2023.107701