Enhancing Quality Control: Defect State Classification of Taralli Biscuits with MobileNet-v2 and DenseNet-201

Industrial production and packaging face significant challenges, such as product damage, color changes, and the presence of foreign bodies. These issues greatly impact product quality, profitability, and marketability, leading to increased consumer complaints. To address these concerns, this study presents a novel method for classifying Taralli biscuits using image processing techniques. The research encompasses a dataset of 4,900 images, featuring four types of defects: no defect, defect-shape, defect-object, and defect-color. Leveraging advanced deep learning architectures, including MobileNet-v2 and DenseNet-201, the classification process achieves impressive accuracy rates of 98.71% and 99.39% respectively. By automating the detection of biscuit damage, the proposed method enhances quality control and inspection processes within the food industry. The combination of state-of-the-art image processing and deep learning techniques in this research provides an effective solution for automatically detecting and categorizing biscuit defects.