ANNs in food
Artificial intelligence (AI), though a popular term nowadays, is not new to food industry. In the late 20’s century, food companies were seeking computational solutions for Quality Control (QC). As product consistency is of great importance to all food companies, standard technical methods are preferable to human testing since it’s subject to individual differences and environmental variations. For that purpose, food scientists were actively working together with engineers and computer scientists for designing and modifying the algorithms. Of all the soft computing techniques, ANN and Fuzzy logic (FL) were the most common approaches, while others, including genetic algorithms (GAs) and support vector machines (SVMs), also had some potentials. For more information please check here (https://www.sciencedirect.com/science/article/pii/B9781845698010500013).
Food is a complex matrix, as we always learnt on any food science classes. Food processing is yet more complex. However, ANN finds itself useful for solving these complicated problems. With the ability to model non-linear relationship between inputs and outputs, ANN could predict without any prior assumptions on the model. One example of its application is in snack frying process, ANN helps to control it based on multiple features such as frying conditions and product properties.
Whether ANN is better than conventional models for solving any practical problems is still under debate, while the consensus is that less underlying information could be drawn from the results of ANN than from the conventional models. The drawbacks of ANN has made it clear that it’s not a replacement of the old causal-effect research approaches, but an alternative for special problems whose variations could not be easily described.
In a review written 10 years ago (https://doi.org/10.1080/10408390600626453), Dr. Huang explicitly summarized the applications of ANN in food science. In the early 21’s century, ANN has already been applied to almost every aspect of food science and technology. In terms of data type for ANN analysis, image, numeric, spectroscopic data are the most common ones. The corresponding sensors for data collection are machine vision, electronic nose/tongue/humans, and spectroscopic equipments such as GC-MS and LS-MS. Not magical at all, ANN is just one approach for statistical analysis whether it’s for regression problems or classification problems.
At the end of the article cited above, the author concluded that ANN’s application in food science, even in varieties of aspect, is still in its developmental stage. Many problems could have been solved better by conventional approaches or other mathematical models due to the scarcity of available data and the complexity of building ANN models. However, it was expected that ANN could be potential as a tool in sensory science. I personally believe that ANN would also be prospective in consumer sciences as human behaviors are extremely complicated and hard to be described by known features.