Page 64 - 2025S

UEC Int’l Mini-Conference No.54 57

Figure 7: User interface: Classification result
Figure 6: User interface: Preprocessing and en- visualization
hancement step

system will be improved to distinguish between
By applying DenseNet201 with transfer learn- dangerous and non-dangerous plants, enhancing
ing and image pre-processing, the system en- its safety for medical and first-aid applications.
ables precise classification of herbal leaf images.
However, the application was built using Python Data availability
and the open-source Streamlit framework, in-
tegrating the DenseNet201 model for reliable, The dataset used in this study is available upon
real-time plant identification. This tool bene- request. Interested researchers may contact the
fits the general public, healthcare professionals,
corresponding author for the access.
and younger users who may struggle to iden-
tify medicinal plants. At present, this system
can recognize only six herb classes. Moreover, Acknowledgments
this application requires a strong internet con-
nection, which may not be available in remote or This study was funded by Shouno Lab. The au-
hilly regions. In future work, we plan to resolve thors are deeply grateful to the lab for provid-
this complexity by developing an Android-based ing computational resources and a supportive re-
offline version of the system to ensure accessi- search environment, as well as for the continuous
bility in any area. Also, expand the dataset to guidance and encouragement that were instru-
include more medicinal plant species and uti- mental in completing this work. The authors
lize advanced AI techniques. Additionally, the also sincerely thank the Japan Student Ser-