DEEP RESIDUAL LEARNING FOR RECOGNISING ADULTERATION IN PULSES VIA THERMAL IMAGING

Abstract

Food is the most essential requirement for sustenance of human life and it is one of the basic necessities of life. For optimal health, we should consume pure, nourishing, and adulterant-free food. Adulteration is the malicious contamination of food products with inferior, less expensive, inedible, or hazardous ingredients. Food adulteration means adding harmful substances to food products to contaminate or adulterate it. Pulses are a significant part of Indian cuisine, serving as a staple food item. Pulses may be adulterated by accidental, commercial, or metallic components, which can be harmful to health, rendering them unsafe for consumption and posing health risks. In order to guarantee the safety and quality of food, it is necessary to analyse pulses for any potential adulteration. In this paper, we propose a method for determining adulteration in Dal (Lentils) using ResNet-50, a deep learning model based on CNN. For this work, thermal images of pure dal and adulterated dal were used to classify the adulterated and unadulterated samples. Experimental results show that the proposed system can achieve high accuracy in detecting adulteration in pulses. The proposed system is also fast and requires minimal human intervention, making it suitable for use in food quality control systems. Key Words: Adulteration, Deep learning, Thermal Images, Convolutional Neural Network (CNN), Residual Network, Pulses, lentils.