Page 63 - 2025S

56 UEC Int’l Mini-Conference No.54

Table 3: Classification Report of DenseNet201 Model

Class Name Precision Recall F1-score Support

Arjun 1.00 1.00 1.00 84
Holy Basil 1.00 1.00 1.00 84
Green Chiretta 1.00 0.99 0.99 84
Malabar Nut 0.99 1.00 0.99 84
Moringa 1.00 1.00 1.00 84
Neem 1.00 1.00 1.00 84

better accuracy than other models. However, eased leaves.
DenseNet201 obtained 99.00% accuracy on the
test set, demonstrating its exceptional perfor- 4.2 Model deployment-MPIcam
mance. While the other four models VGG16 ob-
tained 93.17%, MobileNetV2 obtained 97.99%, For model deployment, the best-performing
ResNet50 obtained 98.59%, and Xception ob- DenseNet201 model was saved in .h5 format and
tained 95.58% test accuracy. This comparison integrated into a Python-based environment us-
of model performances is shown in Table 3. rep- ing Streamlit for real-time interaction. This
resents about the cnn m odel p erformance. lightweight web framework enabled the develop-
ment of a simple and user-friendly interface for
The automated system integrates correctly automated plant identification. To evaluate the
with the DenseNet201 model. Because of model’s predictive capability, we used both the
DenseNet201’s sophisticated CNN design, it training data and a separate set of new images
could capture the characteristics of the ther- for testing. The DenseNet201 model successfully
apeutic plant precisely. Precision and recall identified medicinal plant species, demonstrat-
scores of 1.00 for most classes in the confu- ing its robustness and reliability. Fig.3 presents
sion matrix demonstrated DenseNet201’s reli- tvaluate the efficacy of the trained DenseNet201
able performance. However, the implemented model for automated medicinal plant. identifi-
automated application could help the common cation based on leaf Fig.4 illustrates the flow of
people and the healthcare system to identify the classification process within the application.
accurate Bangladeshi medicinal plants easily. Fig.5 shows about the user interface: Input im-
Users can take an instant photo and identify age selection step. Fig.6 displays User interface:
the plant name along with its facility by us- Classification result visualization. Fig.7 shows
ing the MPIcam application. Additionally, users about the user interface: real time testing visu-
can upload photos from their device gallery to alization. Fig.8 presents about the user inter-
identify medicinal plants. The system deter- face: real time testing result.
mines the relevant plant species by extracting
leaf features and classifying them using pre-
trained models. After post-processing, the out- 5 Conclusion
put is displayed on the screen, and the applica-
tion automatically makes a decision. As a result, This study developed an automated applica-
users can accurately identify medicinal plants tion that can quickly and accurately identify
through leaf feature extraction and obtain re- Bangladeshi medicinal plants without human
liable results with ease, allowing them to utilize effort. We evaluated five deep learning mod-
the plants according to their needs. Future re- els, including VGG16, ResNet50, Xception, Mo-
search can explore larger datasets and alterna- bileNetV2, and DenseNet201, using a primary
tive models to improve the identification process dataset. From the models tested, DenseNet201
as well as distinguish between normal and dis- performed the best with a 99.00% accuracy.