Abstract

The grading of fruits and vegetables is mostly dependent on subjective visual inspection techniques that require labor. Computers and advanced machine learning systems can help to invent non-destructive tools for quality evaluation of fruits and vegetables based on color, size, shape, texture, ripeness, freshness, and surface damage based on computer vision methods. The paper is presenting reviews of methods used for fruit and vegetable grading, including convolutional neural networks, transfer learning, object detection systems, vision transformers, hyperspectral and multimodal vision techniques, and few-shot learning approaches. The review focuses on grading effectiveness, computation complexity, data variety, and algorithms reliability, and it shows that although many studies report the accuracy level higher than 99% while working with particular datasets, most studies used private datasets, making the comparison of results between various studies impossible. In this regard, the paper addresses the issue of a gap between benchmarks and real-life deployment and proposes a framework based on the grading process that includes a confidence lock mechanism for uncertain predictions and a reliable protocol for testing and benchmarking performances of the algorithms This paper lacks any experimental data as the proposed framework and procedures are designed for future assessment.

Keywords

How to Cite This Article

Rights