@@ -557,6 +557,13 @@ Moreover, we recognize that the task itself is inherently challenging due to the
## Conclusion
In conclusion, our project on fruit image classification has demonstrated significant potential in automating the process of identifying fruits based on their images. This aligns well with our initial motivation to enhance the shopping experience in supermarkets, particularly at self-checkout counters. Our results indicate that, with additional resources and time, it is feasible to integrate such a feature into self-service systems.
Although our models do make classification mistakes, their ability to narrow down the possible options for a given fruit image is a step forward in making the user experience more efficient and user-friendly. For instance, rather than scrolling through a long list of fruits, customers could be presented with a few likely options to choose from, based on the system's predictions. This approach not only streamlines the process but also potentially reduces errors and frustration for customers.
Our exploration of different classifiers, from Naive Bayes to CNNs, has provided valuable insights into the capabilities and limitations of machine learning approaches in handling complex image classification tasks. The highest accuracy achieved by the CNN model underscores the effectiveness of deep learning techniques in image recognition, even in scenarios involving subtle distinctions between categories.
Further optimization of the models, exploration of more advanced deep learning architectures, and expansion of the dataset to include a wider variety of fruits and different conditions could enhance the system's accuracy and robustness. These would be interesting options to explore in future work.
