Deep Learning Technique for Automatically Classifying Food Images

Abstract

Because of its growing advantages in the health and medical domains, food image categorization is becoming a more popular study topic.Future diet monitoring systems, calorie estimates, and other projects will undoubtedly benefit from automated food identification techniques. This research presents automated systems for classifying foods using deep learning techniques.The classification of food images using Squeeze Net and VGG-16. These networks are suitable for usage in real-world scenarios in the medical and healthcare industries since it has been shown that employing data augmentation and fine-tuning the hyperparameters significantly improved their performance.Because Squeeze Net is a lightweight network, it is simpler to set up and frequently more appealing. VGG-16 can accomplish quite a decent accuracy even with less parameters. Extracting intricate elements from food photographs allows for even higher categorization accuracy. The suggested VGG-16 network considerably enhances the effectiveness of automated food image categorization. Squeeze Net was suggested as having significantly improved accuracy because of increased network depth.

Squeeze Net performs better in the categorization of food images than VGG-16, according to the results. The name of the food item is categorised with pictures that help you identify it.

With deep learning, larger datasets, and more readily available computer resources, image categorization has become less challenging. The most common and widely applied method for classifying images in the present is the convolution neural network. Various transfer learning algorithms are used to classify images from a broad variety of food datasets. Food is important to life since it gives us various nutrients, thus it's important for everyone to keep an eye on their eating patterns. To live a healthier lifestyle, categorising food is so vital. In this project, pre-trained models are employed rather than the more conventional approach of creating a model from scratch, which reduces computing time and costs while also producing superior outcomes.For training and validation purposes, the food dataset is utilised. It consists of several classes, each with many photos. These pre-trained models will be used to identify the provided food, and they will make predictions about its nutritional value based on the colour of the image.