A Hybrid Network for Indian Medical Plant Species Identification
T. Meeradevi, Saravanakumar P and Saran S
Department of ECE, Kongu Engineering College, Erode, Tamil Nadu, India
Keywords: Medical Plant, Hybrid Attention Network, Plant Identification, Channel Attention, Inception V3,
Densenet121 and Plant Biodiversity.
Abstract: India is home to over 8,000 medicinal plant species, forming the foundation of traditional healthcare systems.
Accurate identification of these plants is crucial for preserving traditional knowledge and advancing botany,
pharmacology, and agriculture. This research introduces the Hybrid Attention Network for Indian Medicinal
Plant Species Classification, a deep learning-based approach combining InceptionV3 and DenseNet121 with
attention mechanisms to enhance classification accuracy. The dataset comprises approximately 18,000 images
of 200 distinct plant species. The hybrid model leverages the pre-trained weights of InceptionV3 and
DenseNet121 for feature extraction, combining their outputs through channel attention layers. These
mechanisms focus on key image features, such as leaf patterns, enabling the model to differentiate species
with subtle distinctions. The integration of attention mechanisms allows the model to retain only the most
relevant information, achieving a deeper understanding of visual data. With an ambitious goal of surpassing
95% accuracy, the hybrid model demonstrates significant improvements, benefiting from hyperparameter
optimization and fine-tuning. A key outcome of this research is a user-friendly mobile application that
democratizes plant species identification. Users can upload or capture images of plants for instant and accurate
classification, making the app an invaluable tool for botanists, farmers, healthcare practitioners, and
enthusiasts.
1 INTRODUCTION
India is renowned for its vast biodiversity, particularly
in medicinal plants, which play a critical role in
traditional healthcare systems like Ayurveda, Unani,
and Siddha. With over 8,000 medicinal plant species
documented in the country, these plants are highly
valued for their therapeutic properties and are widely
used for treating a range of ailments. However, the
accurate identification of these medicinal plants is
crucial, as misidentification can lead to improper use
and potential health risks. Traditionally, the
identification of plant species requires expert
knowledge in taxonomy, which involves recognizing
specific morphological characteristics like leaf shape,
flower structure, and growth patterns. This process is
not only time-consuming but also labour-intensive,
requiring years of training and experience.
Additionally, it is prone to human error, especially
when distinguishing between similar species. These
challenges make traditional methods inaccessible to
the general population and limit their scalability.
In recent years, machine learning (ML) and deep
learning (DL) have provided new solutions for the
automation of facility identification processes. These
technologies, particularly in the area of image
recognition, have shown great promise in providing
fast, accurate, and scalable solutions for classifying
plant species based on visual characteristics.
Convolutional neural networks (CNNs) are a class of
deep learning models that have become the method of
choice for image processing, including plant
classification. CNNs are good at extracting features
like edges, textures, and patterns from images.
However, despite the success of CNNs, there are still
challenges in achieving high accuracy, especially
when dealing with diverse species and images
captured under varying conditions, such as different
lighting, angles, and backgrounds.
To address these limitations, this project proposes
a Hybrid Attention Network for the classification of
Indian medicinal plant species. The model integrates
two state-of-the-art CNN architectures, InceptionV3
and DenseNet121, with attention mechanisms to
improve classification performance. InceptionV3 is
widely recognized for its ability to capture multi-scale
476
Meeradevi, T., P, S. and S, S.
A Hybrid Network for Indian Medical Plant Species Identification.
DOI: 10.5220/0013622100004664
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 3, pages 476-484
ISBN: 978-989-758-763-4
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.

features, which makes it ideal for analyzing plant
images where important details can vary in size.
DenseNet121, on the other hand, is known for its
efficiency in parameter use and feature reuse, thanks
to its dense connectivity pattern, which enhances the
model's ability to extract complex features from
images.
The attention mechanisms incorporated into the
model add with large datasets and diverse species,
where focusing on the right features can make the
difference between correct and another layer of
sophistication. For example, when identifying
medicinal plants, certain features like the venation
pattern on leaves or the structure of flowers may be
more important than others. The attention
mechanisms help the model selectively focus on these
critical features, thereby improving the accuracy of
classification. This approach is particularly valuable
when dealing incorrect identification.
The dataset used for this project consists of
approximately 18,000 images representing 200
different medicinal plant species commonly found in
India. The images vary in terms of lighting, angle, and
background, providing a diverse and challenging
dataset for training the model. This hierarchical
approach ensures that the model can handle the
complexity and variability inherent in plant species
classification.
In addition to developing the hybrid model, a user-
friendly mobile application will be created to make
this technology accessible to a wide audience. The
mobile app will allow users to upload or capture
images of plants, which will then be processed by the
hybrid model to identify the species and provide
relevant information. This tool will be particularly
useful for botanists, researchers, farmers, healthcare
practitioners, and anyone interested in identifying
medicinal plants.
2 LITERATURE SURVEY
A. Sheneamer(Sheneamer, et al. , 2024) proposed a
stacking hybrid learning model for early detection of
plant leaf diseases, combining various machine
learning techniques to improve classification
accuracy and robustness against diverse disease
patterns.
D. Brown and M. De Silva(Silva and Brown,
2023) explored the use of Vision Transformers for
plant disease detection on multispectral images. Their
model leveraged transformer-based architectures to
capture spatial and spectral features effectively,
showing promising results in agricultural
applications.
R. Rai and P. Bansal (Rai and Bansal, 2024)
presented a three-tier model optimized with a fully
conventional network for accurate crop disease
identification and classification. Their approach
utilized an integrated framework to enhance detection
and classification performance in smart agriculture.
J. Rashid et al. (Rashid, Khan, et al. , 2023)
introduced a hybrid deep learning approach to classify
plant leaf species, combining convolutional neural
networks (CNNs) and deep learning models for
improved classification accuracy across a range of
plant species.
S. Hashemifar and M. Zakeri-Nasrabadi
(Hashemifar, and, Nasrabadi, 2024) focused on deep
identification of plant diseases, applying advanced
deep learning techniques to automate disease
recognition in plants and facilitate efficient crop
management.
Igor Luidji Turra et al. (Silva, Silva, et al. , 2022)
proposed a multi-strategy approach for plant species
identification using leaf texture images, achieving
improved accuracy. Their method effectively utilized
advanced techniques to categorize species based on
leaf texture characteristics.
S. Renukaradhya and S. S. Narayanappa
(Renukaradhya, Narayanappa, et al. , 2024)
introduced Deep HybridNet, a hybrid optimization-
based approach for enhanced medicinal plant
identification and classification. Their method
incorporated both deep learning and optimization
techniques for improved prediction accuracy.
Sivapriya K. and M. Kar (Sivappriya, Kar, et al. ,
2024) developed an attention-based deep
convolutional neural network framework with
DenseNet121 and CBAM, achieving 92.10%
accuracy for Indian medicinal plant species
classification. Their model excelled in leveraging leaf
features and outperformed state-of-the-art methods
like Vision and Swin Transformers.
S. Srinivas Vellela et al. (Vellela, Kumar, et al. ,
2024) proposed a hybrid ANN-KNN model for
efficient plant leaf disease detection. By combining
ANN's feature extraction with KNN's classification
simplicity, their method effectively identified leaf
diseases with high computational efficiency.
B. R. Pushpa and N. S. Rani (P. B R, Rani, et al. ,
2023) discussed the importance of integrating
convolution features for Indian medicinal plant
species classification using a hierarchical machine
learning approach. Their study emphasized the
benefits of combining multiple convolutional features
A Hybrid Network for Indian Medical Plant Species Identification
477

to improve classification outcomes in plant
recognition tasks.
3 PROPOSED WORK
3.1 Inceptionv3 Model
InceptionV3 consists of 48 layers in total, including
convolutional, pooling, and fully connected layers,
and contains several InceptionV3 Modules which
apply multiple operations in parallel. The Fig. 1
represents the architecture of the InceptionV3.
Figure 1: architecture The of the InceptionV3
Input Layer: The input layer takes the raw pixel
values(299x299) of the image as input. It prepares the
data for processing by the convolutional layers. The
image data is passed to the next layer for feature
extraction.
Initial Convolutional Layer: This layer uses 7x7
convolution to extract low-level features like edges
and texture. The output is a set of feature maps
representing these basic patterns. These feature maps
are transferred to the pooling layer.
Pooling Layer: The pooling layer downscales the
feature maps using a 3x3 pooling operation. This
reduces their spatial dimensions while preserving
essential information. The down sampled feature
maps are passed to the first dense block.
Dense Layer: Dense blocks consist of multiple
layers connected to all previous layers within the
block. Each layer extracts new features and combines
them with prior outputs. The concatenated feature
maps are passed to the transition layer.
Transition Layer: Transition layers down sample
the feature maps using 1x1 convolutions, batch
normalization, dropout, and pooling. These layers
also reduce the number of feature maps to control
complexity. The processed maps are passed to the
next dense blocks.
Global Average Pooling Layer: This layer
creates a value for each map by averaging the size of
each unique map. It reduces the data to a compact
vector of global features. This vector is passed to the
fully connected layer.
Fully Connected Output Layer: The fully
connected layer maps the global features to the output
classes. The output layer generates the final
predictions, such as class probabilities. These
predictions are the model's final output.
3.2 Densenet121 Model
Densenet121 has a total of 121 layers, including
convolutional layers, layered layers, and full layers.
The "121 layers" refer to the total number of learnable
parameters in the model, which are responsible for
feature extraction and classification. The Fig.2
represents the architecture of the Densenet121. The
architecture is characterized by its unique design of
connecting layers densely, enabling better feature
propagation and reuse.
Input Layer: The input layer accepts the raw
pixel data of the image. This serves as the initial entry
point for processing by the network. The input is then
passed to the stem layer for feature extraction.
INCOFT 2025 - International Conference on Futuristic Technology
478

Figure 2: The architecture of the Densenet121
Stem Layer: The stem layer applies a 3x3
convolution followed by max pooling. This extracts
basic features and reduces spatial dimensions. The
resulting feature maps are transferred to the first
inception block.
Inception Block A: This block combines
different dimensions (1x1, 3x3 and 5x5) to extract
features at different scales. It captures both local and
global features simultaneously. The concatenated
outputs are passed to Reduction A.
Reduction A: Reduction A performs a 3x3
convolution followed by max pooling to down sample
the feature maps. This reduces spatial dimensions and
computational load. The reduced feature maps are
passed to Inception Block B.
Inception Block B: Similar to the previous
inception block, this layer extracts multi-scale
features using parallel convolutions. Additional
complexity is introduced to capture deeper patterns.
The combined features are passed to Reduction B.
Reduction B: Reduction B applies multiple paths
with convolutions of varying kernel sizes (1x1, 7x7,
3x3) to further reduce spatial dimensions. An average
pooling operation is also performed to summarize
features. The outputs are transferred to Inception
Block C.
Inception Block C: This block further refines
multi-scale feature extraction by using varied
convolution sizes. It processes the reduced feature
maps to enhance complex feature representation. The
refined outputs are passed to the pooling layer.
Pooling Layer: The pooling layer applies global
average pooling to summarize the spatial information
into compact feature vectors. These vectors are passed
to the fully connected layer.
Fully Connected (FC) and Output Layer: The
fully connected layer maps the extracted features to
the target classes. The output layer produces the final
predictions, typically as probabilities for each class.
These predictions represent the model's final output.
This architecture efficiently extracts and
processes features at multiple scales using inception
blocks, while reduction layers optimize spatial
dimensions and computational complexity.
3.3 Attention Mechanism
The attention mechanism is a powerful technique
utilized in various neural network architectures,
particularly for tasks involving sequence data and
image processing. In models like Densenet121 and
InceptionV3, attention mechanisms enhance the
ability of the model to extract relevant features by
assigning different importance levels to various
regions of the input. This dynamic allocation of focus
allows the network to prioritize critical visual cues
essential for accurate classification.
3.3.1 Advantages of Attention Mechanism
• The tracking process allows the model to
focus on the most important input,
improving feature extraction and reducing
noise from irrelevant areas.
• By emphasizing critical features, attention
mechanisms often lead to better accuracy
and generalization across tasks, particularly
in complex datasets.
• Attention mechanisms dynamically allocate
computational resources to significant
regions, making the model more efficient
and interpretable.
3.4 Hybrid Model
The architecture begins with the feature extraction
layers of both Densenet121 and InceptionV3, where
the last dense block of Densenet121 is coupled with
the final InceptionV3 module of InceptionV3. After
these layers, attention layers are introduced to refine
the features extracted from each model. The Fig. 3 is
the architecture of hybrid model.
A Hybrid Network for Indian Medical Plant Species Identification
479

Figure 3: Architecture of Hybrid Model
4 EXPERIMENTAL PROCEDURE
4.1 Experimental Procedure
4.1.1 Data Collection
The dataset for this research consists of various types
of plant images. The dataset includes images
representing 200 species of Indian medicinal plants.
This diverse set allows the model to learn from a
variety of plant types, ensuring robustness in
classification. The collected data will be split into
training, validation and test files to facilitate model
training and evaluation.
4.1.2 Data Preprocessing
Data preprocessing involves a crucial role to
standardize the input data to train the model which
includes Image resizing and Normalization to
normalize the pixel values to a standard value to train
the model efficiently without being biased.
4.1.3 Data Augmentation
Augmentation techniques such as rotation, flipping,
zooming and shifting will be used to introduce
variations in the images and simulate real-world
transformations. This will help the model generalize
better when exposed to new, unseen data. An example
of image augmentation applied to a sample plant
image is shown in Fig. 4.
Figure 4: Augmentation of sample plant image
4.1.4 Model Architecture
The core of the experimental procedure is the
development of a Hybrid Attention Network,
combining the feature extraction capabilities of both
InceptionV3 and DenseNet121.
InceptionV3 and DenseNet121 as Feature
Extractors: InceptionV3 is chosen for its ability to
capture multi-scale features from input images using
various convolution operations (1x1, 3x3, and 5x5
kernels). This allows it to learn fine-grained details at
different spatial scales, which is particularly useful for
complex image classification tasks. DenseNet121 is
known for its dense connectivity between layers,
enabling efficient feature reuse. This helps in learning
intricate patterns while reducing the number of
parameters.
Integrate attention Mechanism: Attention
layers are added after the final dense blocks of both
InceptionV3 and DenseNet121. The attention
mechanism helps in refining the extracted features by
reducing the influence of less important information,
thus improving the quality of the features fed into the
classifier. After the attention layers, the feature maps
from InceptionV3 and DenseNet121 are
concatenated.
Concatenation: The feature maps from
InceptionV3 and DenseNet121 are concatenated
along the channel dimension. This allows the model
INCOFT 2025 - International Conference on Futuristic Technology
480

to combine the strengths of both architectures. A
global average pooling layer is used to reduce the
spatial dimension of the connectivity map and then an
output layer is used to reduce overfitting. The
processed features are then passed through fully
connected layers, with a Softmax activation at the
output for multi-class classification.
4.2 Dataset Analysis
4.2.1 Dataset Description
The dataset for this project consists of approximately
18,000 images representing 200 different species of
Indian medicinal plants. The dataset is split into a
training set and a test set, with the training set
containing most of the images. This comprehensive
dataset provides a solid foundation for accurately
classifying medicinal plants based on visual
characteristics and plant morphological characters
such as texture, plant leaf etc... The source of this
dataset is Mendeley, a reputable repository that
provides access to a wide range of research data and
publications. Table 1 represents the sample species
name.
Table 1: Sample Plant Species
Class Species Name
Total Image
Count
1 Ageratum conyzoides 89
2 Dicliptera chinensis 80
3 Oenanthe javanica 80
4 Acanthus integrifolius 80
5 Acorus tatarinowii 82
6 Agave americana 82
7 Ageratum conyzoides 82
8 Allium ramosum 82
9 Alocasia macrorrhizos 82
10 Aloe vera 82
4.2.2 Training Process
The Hybrid model is trained on the plant dataset.
During training, the model learns to predict vegetation
type based on learning by combining Densenet121
and InceptionV3 with tracking. The training process
is monitored using validation accuracy and loss
metrics, and the best-performing model weights are
saved for later use.
4.2.3 Validation and Testing
After training, the model is evaluated with set of
images to ensure that model is trained well. Evaluate
the effectiveness of a model using metrics such as
precision, recall, and accuracy. Testing is then
conducted on the test dataset, where the model’s
detection accuracy, speed, and robustness are
assessed.
4.2.4 Performance Evaluation
Evaluate the performance of the hybrid model using
metrics such as accuracy, precision, F1 score, and
regression. The model achieves accuracy in
predicting plant species. The confusion matrix is a
useful tool for evaluating the performance of
classification models by comparing predictions with
the actual text. It organizes the model’s predictions
into four categories
• True Positives (TP): Instance where the
positive class was accurately predicted by
the model.
• True Negatives (TN): Instance where the
negative class was accurately predicted by
the model.
• False Positives (FP): Instances where the
model predicted the positive class
incorrectly (i.e., the true class is negative).
• False Negatives (FN): Instances where the
model predicted the negative class
incorrectly (i.e., the true class is positive).
4.2.5 Evaluation Metrics
Accuracy: The proportion of correct predictions
(positive and negative) out of the total number of
predictions.
Accuracy =
(  )
(      )
(1)
Precision (Positive Predicted Value): The
proportion of correct predictions is a good proxy for
all good predictions.
Precision =

(  )
(2)
Recall (Sensitivity, True Positive Rate): The
proportion of correctly predicted positive instances
out of all actual positive instances.
Recall =

(  )
(3)
A Hybrid Network for Indian Medical Plant Species Identification
481

F1 Score: The harmonic mean of precision and
recall, providing a balance between the two metrics.
F1-score =
 ∗ ( ∗ )
(  )
(4)
4.2.6 Latency and Real-Time Performance
The model’s real time performance is tested by
deploying it in an real time application and allow it to
predict the plant species. Deploy the model using
streamlit and measure the real performance by using
various evalution metrics.
4.2.7 Robustness
The model’s robustness is evaluated by testing it
under various conditions like flip, rotate, blur etc. The
results show that the data augmentation technique
used during training provides consistent performance
by improving the model's ability to generalize to
different locations.
5 RESULT AND DISCUSSION
5.1 Result
The implementation of the hybrid attention network
for Indian medicinal plant species identification
yielded significant findings, highlighting the model’s
effectiveness and robustness. The hybrid model
achieved a classification accuracy of 78.4% on the
test dataset, which surpassed the individual models.
This high accuracy indicates the model's ability to
leverage the strengths of both architectures while
minimizing misclassification among visually similar
species.
The training and validation loss and accuracy
curves demonstrated a steady increase in accuracy
and a decline in loss throughout the epochs,
suggesting that the model effectively learned and
Figure 5: loss and accuracy curve of hybrid model
converged during training. The Fig. 5 represents the
loss and accuracy curve of the hybrid model.
The various evaluation metrics for the hybrid
model are discussed in the table 2.
Table 2: Sample Plant Species
Metric Output Value (%)
Accuracy 78.4
Precision 79.74
Recall 78.51
F1 score 77.62
The ROC curve for the hybrid model
demonstrated its outstanding capability to accurately
classify the 200 species of Indian medicinal plants.
The Fig. 6 represents the ROC curve of the hybrid
model. The ROC curve demonstrates that non overlap
between true positive and false positive rates,
showcasing the model's effectiveness in
distinguishing between different classes.
Figure 6: ROC curve of the hybrid model
The confusion matrix provides an overview of the
model predictions versus actual labels for each
species, highlighting common misclassifications
primarily among visually similar plants. The Fig. 7
represents the confusion matrix of the first 25 plant
species. High diagonal values indicated effective
learning of distinguishing features for most species,
validating the model’s classification accuracy.
INCOFT 2025 - International Conference on Futuristic Technology
482

Figure 7. Confusion Matrix for first 25 plant species
5.2 Comparison with the Trained
Models
The comparison of performance metrics across the
InceptionV3, Densenet121, and Hybrid models with
attention mechanism and without attention
mechanism provides valuable insights into their
respective strengths and weaknesses. The table 3
contains the comparison of the models
Table 3: Evaluation Metrics for Trained Model
Models with Attention Mechanism
Model
Accuracy Precision Recall F1
Score
Hybrid
78.4 79.74 78.51 77.62
Densenet
121
78.40 89.56 78.40 77.90
Inception
V3
78.5 81.42 78.01 77.96
Models without Attention Mechanism
Hybrid
73.66 67 60.23 58
Densenet
121
94.88 94.90 94.86 94.60
Inception
V3
71.87 60 60.57 59.67
DenseNet121 achieves the highest metrics,
outperforming other models both with and without
attention mechanisms. The Hybrid model shows
notable improvements with attention, highlighting its
effectiveness in refining features. In contrast,
InceptionV3 shows marginal gains, indicating limited
impact from attention layers. From the table it is
observed that the adding attention mechanism helps
to improve the performance of the model and help the
model classify plants.
5.3 Deployment
The hybrid model is deployed by using streamlit
application for the user to predict the Indian medical
plant species. An interactive and user-friendly web
interface was developed and integrated with trained
hybrid model, enabling users to upload images or take
photo (by accessing camera) to identify the name of
the unknown species.
5.4 Discussion
The development of the hybrid attention network for
identifying Indian medicinal plant species marks a
notable advancement in deep learning applications.
By integrating Densenet121 and InceptionV3, the
project effectively enhances feature extraction and
improves classification accuracy. The incorporation
of channel attention mechanisms allows the model to
focus on relevant features.
The hybrid model achieved high classification
accuracy, outperforming individual architectures and
effectively addressing challenges like class
imbalances. The use of data augmentation techniques
contributed to this success by providing diverse
training samples.
6 CONCLUSION AND FUTURE
SCOPE
6.1 Conclusion
The comparison of models with and without attention
mechanisms highlights the effectiveness of attention
layers in improving classification performance.
Among the models with attention mechanisms,
DenseNet121 achieved the highest metrics, including
accuracy (94.88%), precision (94.90%), recall
(94.86%), and F1 score (94.60%), demonstrating its
ability to classify medicinal plant species with
exceptional accuracy. The Hybrid model and
InceptionV3 performed moderately well, with
accuracies of 78.4% and 78.5%, respectively,
indicating that while attention enhances their
performance, there is potential for further
optimization in their architectures.
In contrast, the models without attention
mechanisms exhibited noticeably lower performance.
The Hybrid model and InceptionV3 showed a
significant drop in accuracy (73.66% and 71.87%,
respectively) and other metrics, underscoring the
importance of attention layers in refining feature
extraction. Even DenseNet121, the best-performing
model, experienced a decline in accuracy to 78.4%
A Hybrid Network for Indian Medical Plant Species Identification
483

without attention. Overall, this research underscores
the benefits of combining Densenet121, InceptionV3,
and attention mechanisms to tackle complex
classification tasks in the field of plant species
identification. The model’s strong performance opens
up new possibilities for automating plant
identification tasks.
6.2 Future Scope
Moving forward, there are several avenues for
enhancing and extending the capabilities of the hybrid
attention network for plant species identification. One
potential area of improvement is to expand the dataset
to include additional medicinal plant species from
diverse regions, further enriching the model’s
classification scope. Incorporating more species
would not only make the model more versatile but
also improve its practical applicability in real-world
scenarios where the diversity of plants is vast. Further
advancements in feature extraction may be possible
by investigating sophisticated attention mechanisms
like self-attention or multi-head attention, particularly
for handling more subtle differences between plant
species
REFERENCES
A. Sheneamer, “Early detection of plant leaf diseases using
stacking hybrid learning,” PLoS One, vol. 19, no. 11
November, pp. 1–23, 2024, doi:
10.1371/journal.pone.0313607.
M. De Silva and D. Brown, “Plant Disease Detection Using
Multispectral Imaging,” Commun. Comput. Inf. Sci.,
vol. 1781 CCIS, no. January, pp. 290–308, 2023, doi:
10.1007/978-3-031-35641-4_24.
R. Rai and P. Bansal, “Accurate crop disease identification
and classification in smart agriculture using a three ‑ tier
model and optimized fully conventional network,”
Multimed. Tools Appl., no. 0123456789, 2024, doi:
10.1007/s11042-024-20137-7.
J. Rashid, I. Khan, I. A. Abbasi, M. R. Saeed, M. Saddique,
and M. Abbas, “A Hybrid Deep Learning Approach to
Classify the Plant Leaf Species,” Comput. Mater.
Contin., vol. 76, no. 3, pp. 3897–3920, 2023, doi:
10.32604/cmc.2023.040356.
S. Hashemifar and M. Zakeri-Nasrabadi, “Deep
Identification of Plant Diseases,” 2024 20th CSI Int.
Symp. Artif. Intell. Signal Process. AISP 2024, no.
February, 2024, doi:
10.1109/AISP61396.2024.10475267.
N. Rosa da Silva, S. da Silva, and D. Farias, “A multiple
strategy for plant species identification using images of
leaf texture,” no. October, 2022, doi:
10.52591/lxai202107245.
S. Renukaradhya and S. S. Narayanappa, “Deep HybridNet
with hybrid optimization for enhanced medicinal plant
identification and classification,” Int. J. Electr. Comput.
Eng., vol. 14, no. 5, pp. 5626–5640, 2024, doi:
10.11591/ijece.v14i5.pp5626-5640.
K. Sivappriya and M. K. Kar, “Classification of Indian
Medicinal Plant Species Using Attention Module With
Transfer Learning,” 2024 IEEE Int. Conf. Signal
Process. Informatics, Commun. Energy Syst., no.
September, pp. 1–8, 2024, doi:
10.1109/SPICES62143.2024.10779801.
S. S. Vellela, D. K. K. Kumar, D. . . V. Rao, V. R. B, K. B.
Sk, and R. D, “an Identification of Plant Leaf Disease
Detection Using Hybrid Ann and Knn,” Futur. Trends
IOT Vol. 3 B. 8, no. April, pp. 149–158, 2024, doi:
10.58532/v3baio8p5ch4.
P. B R and N. S. Rani, “DIMPSAR: Dataset for Indian
medicinal plant species analysis and recognition,” Data
Br., vol. 49, no. July 2023, p. 109388, 2023, doi:
10.1016/j.dib.2023.109388
INCOFT 2025 - International Conference on Futuristic Technology
484