Recognition of inlet wet food in drying process through a deep learning approach

Summary

Smart drying is one of the newest and most promising techniques. It is a multi- and inter-disciplinary sector which has potential to guarantee high value end-products by implementing innovative and reliable sensors, resources, tools and practices. Its recent developments embrace various R&D areas, such as computer vision (CV) and deep learning, which deal with allowing computers to understand digital images and videos better than humans. Conventional machine-learning techniques suffer several limitations, mainly due to their inability to process raw data. In fact, in the last few decades, machine learning required considerable domain expertise to mine raw data and extract features from which an algorithm could identify patterns in the input. Deep learning is a novel subfield of machine learning, which embraces methods that allow to discover patterns for detection or classification purposes by using raw data. Consequently, CV in combination with deep learning has the potential to be a powerful Process Analytical Technology tool useful for enhancing the understanding and control of critical process parameters that impact on quality of the final product.
Deep learning was tested for its feasibility as CV tool for the analysis of inlet wet food to drying process. In details, convolutional neural networks (CNNs) were successfully applied for addressing the following tasks: (i) the semantic image segmentation of the inlet product (i.e., recognition between background and product pixels); (ii) the inlet product classification through its segmented image; (iii) the automated selection of optimal settings of drying process parameters.
Results obtained not only represent a step forward in the development of smart dryers able to recognise the inlet wet product, and to set the proper process parameters on its own or as decision support system, but also lay the foundation for further researches on using a computer vision system as PAT tool for smart drying processes.