Transfer Learning Based Models for Food Detection Using ResNet-50

Abstract

Being overweight may be caused by eating too many calories. It is a curable medical condition defined by abnormal fat accumulation in the body. Diabetes, excessive cholesterol, and heart attacks are the most common, although high blood pressure, colon cancer, and prostate cancer are also common. Computer techniques are often utilized to address such difficulties. In this work, we develop a system that detects and identifies food allergies using food photographs. To summaries, powerful computer algorithms such as transfer learning (ResNet50) have been taught to detect food type and validate the identified label in dataset food 101, as well as supply nutrients. The fundamental purpose of this study was to create a single framework capable of managing the difficult process of detecting, localizing, and classifying food allergies. Furthermore, larger weight parameter optimization using Adam and RMS Prop optimizers was attempted to increase their performance on healthy and allergic food image datasets. The Resnet-50 was trained to obtain the greatest mean average accuracy when compared to the other transfer learning meta-architectures. It achieved the best-identifying results by utilizing an Adam optimizer and obtaining 95% accuracy. The suggested technique was discovered to be novel since it detects all food types and then provides the nutrients of that meal from another dataset. In reality, employing the transfer learning technique to successfully diagnose food allergies would assist to prevent the adverse application of issues in diet management.