Page 92 - IJAMD-2-2

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

Figure 10. Results obtained from different feature metrics evaluated on
the system

The results indicate that the SVM model performed
Figure 8. Scatter plot of fruit classification between mango and tomato well in classifying the fruits, achieving high scores across
all metrics. This is attributed to the effective feature
extraction, the selection of an optimal kernel, and careful
tuning of the C value.
Therefore, the performance evaluation has identified
the capability of the SVM model to classify fruits with
reasonably high accuracy using only visual features. This
reinforces the potential of image-processing techniques in
agricultural applications. The findings also emphasize the
importance of selecting appropriate kernels, optimizing
model parameters, and use of comprehensive feature sets
to further improve the classification accuracy of the system.
4. Discussion

The use of image processing techniques in fruit
classification solves all the problems associated with the
traditional method of classification, such as subjectivity
Figure 9. Scatter plot of fruit classification between mango and orange
in classification and potential damage to the fruit. Various
features can be utilized in the classification system; while
SVM model efficiently captures the underlying patterns in some features allow highlighting the differences across
the data, enabling accurate classification.
fruit classes easily and accurately, the selection of the most
Figure 9 provides another visualization of hyperplanes impactful features helps reduce the computational power
in a two-dimensional feature space. In contrast to Figure 8, required.
where there is a distinct separation between mango and
tomato based on the color feature but not size, resulting There are various machine learning models available.
46
in a vertical hyperplane. Figure 9 shows a clear separation Naskar and Bhattacharya utilized artificial neural
between mango and orange. Here, both color and size networks and achieved an accuracy above 90%. Similarly,
47
features contribute to the distinction, resulting in a Bahaghighat et al. proposed the use of neural networks
47
diagonal hyperplane. and reported reasonable results. Bahaghighat et al.
demonstrate that the SVM model is comparable in
3.7. Performance of the SVM model effectiveness to other models while offering the advantage
The performance of the SVM model was evaluated using of requiring a significantly smaller dataset.
several standard metrics, as depicted in Figure 10, which The SVMs are primarily designed for linearly separable
shows the model’s accuracy, precision, recall, and F1 score data. However, when dealing with non-linear data, a kernel
in the context of fruit classification. function can be applied. The kernel effectively projects the

Volume 2 Issue 2 (2025) 86 doi: 10.36922/IJAMD025150011