Research on YOLOv5 algorithm and its application for accurate and rapid detection of ochratoxin A in agricultural samples

Authors:
  • MSMingNa Sun
    MingNa Sun
  • GWGuizhu Wu
    Guizhu Wu
  • JFJiayun Fu
    Jiayun Fu
  • JZJingjing Zhong
    Jingjing Zhong
  • GLGuoxing Liu
    Guoxing Liu
  • JDJinsheng Duan
    Jinsheng Duan
Preprint
Publisher website
Google Scholar
Abstract
Ochratoxin A (OTA) is a highly toxic mycotoxin prevalent in agricultural samples. In order to overcome the limitations of traditional detection methods, this study developed a rapid visual detection system based on deep learning by optimizing the YOLOv5 algorithm. Through integrating attention mechanism to enhance color feature extraction and employing dilated convolutions to expand the receptive field, the proposed Efficient-YOLOv5 algorithm significantly improved the accuracy of OTA concentration identification. Comparative experiments demonstrated the superior performance of Efficient-YOLOv5, with mean average precision (mAP) increasing from 92.86% to 97.65%, precision from 93.34% to 95.73%, recall from 91.67% to 92.68%, and F1-score from 90.34% to 91.69%. Validation on real agricultural samples further confirmed the high accuracy of the method, yielding a detection error of 0.04 compared to standard concentration and 0.25% under simulated scenarios. The optimized loss function achieved a mAP value of 91.37%, precision of 91.86%, and F1-score of 88.26%, highlighting its efficiency, accuracy and high-throughput potential. By combining visual data with deep learning, the proposed detection system offered a promising framework for the rapid, accurate and efficient identification and detection of OTA.

This publication has 16 references indexed in Scilit: