Food Calorie Recommender System Using Mobilenet Deep Learning Technique
Keywords:
Food 101 dataset, Food classification, Deep Learning, Transfer Learning, image processing, CNN, VGG-16, SqueezenetAbstract
Food image classification has gained significant attention in recent years due to its relevance in health monitoring, dietary assessment, and medical applications. Automated recognition of food items from images can assist in calorie estimation, nutrition tracking, and disease-specific diet planning. This study presents a deep learning-based approach using Convolutional Neural Networks (CNNs) for classifying food images. Two widely used models, SqueezeNet and VGG-16, are implemented and evaluated. Performance enhancement is achieved through data augmentation techniques and hyperparameter tuning. SqueezeNet, being lightweight and efficient, is ideal for real-time deployment on resource-constrained devices. Meanwhile, VGG-16, despite its larger size, demonstrates strong accuracy due to its deep architecture. Complex feature extraction further boosts classification performance in both models. Experimental results indicate that the proposed fine-tuned SqueezeNet achieved higher classification accuracy compared to the baseline VGG-16. These findings support the potential of deploying deep learning models for practical food image recognition in healthcare applications.
Downloads
References
Zhou et al. (2019) provided a comprehensive overview of how deep learning techniques have been applied in various aspects of food-related analysis, highlighting its effectiveness in enhancing food quality assessment and classification tasks.
Farinella, Moltisanti, and Battiato (2014) explored a method for food image classification by utilizing a Bag of Textons representation, presenting their findings at the IEEE ICIP conference in Paris.
Zhou, Lapedriza, Xiao, Torralba, and Oliva (2014) discussed a strategy for scene recognition using deep features extracted from the Places dataset, demonstrating its application at the NIPS conference.
Rahmani (2017) introduced a novel spectral clustering method that integrates both color and texture features, significantly improving segmentation accuracy in PolSAR images, as published in National Academy Science Letters.
Wang et al. (2017) proposed a classification model for remote sensing imagery that combines an optimized support vector machine with a modified ant colony optimization algorithm, detailed in the Information Sciences journal.
Xia, Ghamisi, Yokoya, and Iwasaki (2018) suggested the use of random forest ensembles coupled with multiextinction profiles for effective classification of hyperspectral images, as featured in IEEE Transactions on Geoscience and Remote Sensing.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 International Journal of Scientific Research in Science, Engineering and Technology
This work is licensed under a Creative Commons Attribution 4.0 International License.
