Milk Box Defect Detection using Deep CNN Model

Abstract

Food packaging is a crucial issue in the food industry. In the milk and dairy industries, milk boxes must be in a good condition to preserve the freshness and quality of the milk. Any defect on the box is not compromised because milk will be spoiled and cause harm to the consumers. This paper focuses on the use of the deep CNN model in detecting the defect in the milk box. The work employed a pre-trained AlexNet model for the two-class classification problem, PERFECT and DEFECT. This work also aims to evaluate the box defect detection accuracy in terms of percentage through experiments on static images. 80 milk box images were used in this preliminary study, in which 40 images of perfect boxes and 40 images of the defect boxes. The result that is obtained by this project is 100% accuracy. It is concluded that the trained model can perform relatively well in classifying milk box images. However, in the future, more data could be used in the training to increase the accuracy of the model and avoid the occurrence of overfitting in the model. Further possible attempt also includes the use of milk box images from other imaging technologies in the training to give higher variability on the choice of dataset using which a deep learning model can train.