Deep learning-driven Vis/NIR spectroscopic devices for fruit quality assessment: A comprehensive review

Abstract

Background

The increasing availability of visible and near-infrared (Vis/NIR) spectroscopic devices has accelerated their widespread application in fruit quality assessment. However, traditional machine learning methods face challenges in processing the complex and unstructured data generated by these devices, often encountering the generalization bottleneck in intricate application scenarios. Recently, deep learning (DL) has yielded remarkable advancements in the field of spectral analysis, enhancing the accuracy of fruit quality assessment and offering promise opportunities for continued innovation in the field.

Scope and approach

This paper presents a comprehensive overview of DL-based Vis/NIR spectroscopic devices for fruit quality assessment. It introduces the key components, advantages, disadvantages, and application scenarios of various Vis/NIR spectroscopic devices, along with relevant DL algorithms and architectures. Furthermore, the integration of DL with spectroscopic devices for fruit quality assessment is summarized, with an emphasis on the contributions and limitations of different DL architectures. Current challenges and future research directions are also discussed.

Key findings and conclusions

DL-based Vis/NIR spectroscopic devices demonstrate strong potential for fruit quality assessment. To optimize cost-effectiveness and accuracy, device selection and DL architecture design should be tailored to specific application scenarios. Although some limitations persist, ongoing technological innovations are expected to drive the broader industrial and commercial adoption of DL-enabled Vis/NIR spectroscopic systems, ensuring improved fruit quality and authenticity in the future.