Page 93 - IJAMD-2-2

International Journal of AI for
Materials and Design Fruit image detection using AI

input data into a higher-dimensional feature space, where with mechanical sorting provides a proof of concept for a
the data are considered linearly separable. 48 fully automated fruit classification and sorting system.
According to Yekkehkhany et al., the linear kernel The image processing system developed in the study
48
is the simplest and fastest to process, but the polynomial demonstrates a reliable and efficient approach to fruit
kernel, though more time-consuming and complicated, classification, with the potential to enhance quality control
often yields more accurate and reliable results. In this and reduce labor costs in the agricultural sector. The findings
study, the polynomial kernel achieved a much greater further emphasize the potential of advanced technologies to
accuracy than a linear kernel. This suggests that when replace or improve traditional fruit classification methods.
more than two features are used for fruit classification, Future research should focus on extending the system
utilizing the linear kernel may negatively affect the results to accommodate a wider variety of fruits, along with the
obtained. integration of machine learning algorithms for real-time
Color features played a significant role in the fruit classification and sorting.
classification of the fruits. As such, the quality of the Acknowledgments
camera and the lighting conditions during image capture
are crucial. Poor image quality or inconsistent lighting None.
may distort color representation, particularly due to
background interference, which may affect classification Funding
results. None.

5. Conclusion Conflict of interest
This paper presents an image processing system developed The authors declare that they have no competing interests.
for classifying selected fruits, including oranges,
tomatoes, and mangoes. This system demonstrated Author contributions
tremendous potential in enhancing both the efficiency Conceptualization: All authors
and accuracy of fruit sorting within associated processes. Data curation: Oluwaseun Emmanuel Oyewande, John
The system achieved impressive performance metrics, Audu
with an accuracy of 100%, precision of 96%, recall of Formal analysis: All authors
92%, and F1 score of 89%, as shown in Figure 10. These Investigation: Oluwaseun Emmanuel Oyewande
results demonstrate the effectiveness of the implemented Methodology: All authors
techniques and the robustness of the model in accurately Project administration: Babatunde Olayinka Oyefeso, John
classifying fruits.
Audu
The findings highlight the importance of feature Resources: All authors
selection in the classification process. With the inclusion Software: Oluwaseun Emmanuel Oyewande, John Audu
of multiple features, the accuracy of the system increased Supervision: Babatunde Olayinka Oyefeso, John Audu
significantly. This indicates that certain features contribute Validation: Babatunde Olayinka Oyefeso, John Audu
more strongly to overall classification performance Visualization: Oluwaseun Emmanuel Oyewande, John
compared to others. For the SVM model, various kernel Audu
functions were evaluated. The results show that the Writing – original draft: Oluwaseun Emmanuel Oyewande
polynomial kernel outperformed both the linear kernel Writing – review & editing: John Audu
and the RBF kernel, demonstrating its effectiveness in
handling non-linear relationships within the data. Ethics approval and consent to participate
Furthermore, it is essential to optimize the C value. This Not applicable.
study suggests that the optimal C value is tightly connected
with the various features’ values. This indicates the need Consent for publication
for dataset-specific parameter tuning to achieve optimal Not applicable.
performance.
Availability of data
The integration of the Arduino board with the
automated sorting mechanism, guided by the classification The code, analysis scripts, and datasets supporting this
results from the SVM model, sorted all fruits into their article have been included as part of the supplementary
respective containers. This integration of image processing material.

Volume 2 Issue 2 (2025) 87 doi: 10.36922/IJAMD025150011