A hybrid sensor and CNN based system for formalin detection and quality classification of apples

Abstract

Food adulteration poses a serious health threat in many developing countries. One of the most alarming practices is the use of formalin to prolong the shelf life of fruits, especially apples. This paper introduces a system that combines formaldehyde gas detection sensors with image-based classification to identify formalin adulteration and categorize apples by quality. Its main idea is the immediate detection of formalin using a threshold value of around 2-3.9 volts with an MS1100 semiconductor gas sensor that will detect formalin for any value above the threshold value, while the assessment of fruit quality is in turn done by a convolutional neural network (CNN) which was expected to group the images with over 98% accuracy and below 5% recall. The CNN model consists of multiple convolutional layers that act on the apple images captured using the webcam for feature extraction. The results show that, although the formalin detection system is very accurate while the quality classification model looks promising.