Deep learning neural networks for acrylamide identification in potato chips using transfer learning approach

Acrylamide is a carcinogenic chemical compound found in carbohydrate rich foods when fried and baked at high temperatures, like potato chips. Identification of such toxic substances in food items is of tremendous significance. Conventional identification approaches like liquid chromatography-mass spectrometry (LC-MS) are time-consuming, destructive and require trained manpower. Traditional machine learning methods involve the extraction of handcrafted features that needs to be judiciously selected. To overcome such shortcomings of the existing researches, an alternate method incorporating deep convolutional neural network (DCNN) for acrylamide identification has been proposed. The novelty of the proposed research work provides an opportunity to explore and distinguish between traditional machine learning and deep learning techniques. Also, the novel contribution in the proposed research work remarkably improves computation complexity which thereby, increases its system accuracy. Deep learning models, pre-trained on ImageNet dataset, showed a remarkable performance in comparison to existing methods. Simulation results demonstrate that MobileNetv2 out-performed AlexNet, ResNet-34, ResNet-101, VGG-16 and VGG-19 models. Therefore, the vitality of algorithm used, validates the advantages of the proposed research work, which could be used as an efficient and effective tool for food-quality evaluation in real-time applications.