Multispectral fusion and lightweight CNN model framework for nondestructive detection of grain quality

Background: The detection of grain quality has always been an important task in ensuring food safety. Traditional detection methods often require complex equipment or manual operations, which are not only inefficient but also costly and prone to damaging the grain. With the development of technology, non-destructive detection methods based on image processing and deep learning have gradually emerged as a new direction for solving these problems. Objective: This study aims to propose a framework for non-destructive detection of grain quality based on multispectral fusion and a lightweight CNN model to improve detection efficiency, reduce costs, and enhance detection accuracy. Methods: This study uses multispectral image data for fusion processing, employing a lightweight CNN model for image classification and quality assessment. The effectiveness of the combination of multispectral fusion and the CNN model in non-destructive grain quality detection is evaluated. Results: Experimental results show that the proposed multispectral fusion and lightweight CNN model can significantly improve the accuracy and efficiency of grain quality detection. Although the lightweight CNN slightly lags behind YOLOv5s and CNN in terms of performance, it still demonstrates good performance, especially under Fused Image processing. Despite an accuracy of 84.16%, its computation speed and resource consumption are significantly lower than other models. Innovation: The innovation of this study lies in the proposed non-destructive detection framework that combines multispectral imagery with a lightweight CNN, integrating the fusion of multispectral images and the lightweight design of CNN models, thereby significantly improving the real-time performance and accuracy of grain quality detection.