Food defect detection technologies based on deep learning and prospects in detection of unsound wheat kernels

With rising concerns over global food security and quality pressures and the rapid advancement of agricultural intelligence, wheat quality detection demands higher efficiency, accuracy, and automation. Unsound wheat kernels, which adversely affect flour yield, storage stability, and food safety, have become key factors in grain processing and storage. Traditional detection methods are constrained by low efficiency, poor consistency, and limited real-time capabilities, rendering them inadequate for the high-throughput, non-destructive requirements of modern grain industry chains. Recent advances in machine vision and deep learning have facilitated novel approaches for the precise recognition and intelligent sorting of unsound wheat kernels. This paper provides a comprehensive review of the current research status and technological progress in this domain, focusing on image acquisition system, feature extraction, model optimization, multimodal fusion, and lightweight deployment. Furthermore, the advantages and challenges of end-to-end automated detection systems in practical applications are critically examined. This study aims to offer theoretical insights and methodological guidance for advancing intelligent wheat quality detection technologies, thereby supporting the broader implementation of smart agriculture practices.