For now our system perform well on test dataset. However, while applying our system on real-time image, the image recognition might fail due to the variant circumstance. Many factors may influence the performance of our models. For example, in different brightness circumstance, the color of the fruits might seems different in pictures. Also, we found it hard to distinguish fruits with similar appearance and different size (Orange and grapefruit for example). To solve this problem we might need more robust model. We should augment the training data, apply stronger deep learning models such as VGG or AlexNet.
Besides that, our system can not keep track of every fruit. We only record fruits bought in same date and user have to input the amount manually. To solve this problem, we should consider semantic segmentation instead of image recognition, and perform real-time monitor on video stream in refrigerator.
Besides that, our system can not keep track of every fruit. We only record fruits bought in same date and user have to input the amount manually. To solve this problem, we should consider semantic segmentation instead of image recognition, and perform real-time monitor on video stream in refrigerator.