FOOD SCIENCE ›› 2022, Vol. 43 ›› Issue (23): 41-48.doi: 10.7506/spkx1002-6630-20211125-312

• Basic Research • Previous Articles     Next Articles

Machine Learning Combined with Multi-Feature Fusion of Vision and Tactile Sensation for Quality Grade Evaluation of Salted Sea Cucumber

ZHU Xinyu, KANG Jiaming, SHAO Weidong, LIU Yang, ZHANG Xu, YANG Jixin, WANG Huihui   

  1. (1. School of Mechanical Engineering & Automation, Dalian Polytechnic University, Dalian 116034, China;2. National Engineering Research Center of Seafood, Dalian 116034, China;3. Cross Science of Food, Dalian Polytechnic University, Dalian 116034, China)
  • Online:2022-12-15 Published:2022-12-28

Abstract: The results of quality grade evaluation of salted sea cucumber have a direct effect on its economic value. It is difficult to distinguish different grades of salted sea cucumber according to their color, shape and smell. The existing detection methods, which are characterized by offline decentralized operation and dependence on manual experience, cannot meet the requirements of large-scale production lines. A new multi-sensory feature fusion method based on machine visual and mechanical features was proposed for rapid and nondestructive quality grade evaluation of salted sea cucumber in this study. In view of the complex relationship between the salt content and texture properties of sea cucumber, the mechanical information and deformation images of sea cucumber during the stress recovery process were perceived by the quality grade evaluation system developed. The quality grade evaluation of salted sea cucumber was realized by using machine learning algorithm. In order to accurately extract image features related to the change of sea cucumber contours, the traditional image processing method was improved to establish a dynamic energy map for sea cucumber contours. Then, texture features were extracted by histogram (HIS) and gray-level co-occurrence matrix (GLCM). For mechanical information, mechanical statistical features were extracted. All feature data were subjected to dimensionality reduction and fusion by one-way analysis of variance (ANOVA) and principal component analysis (PCA) to obtain the multi-feature fusion of vision and tactile sensation. According to the characteristics of the fused data, the genetic algorithm (GA) was used to optimize the support vector machine (SVM) to establish a grade evaluation model. It was shown that the SVM model had excellent performance, with Accuracy = 1, Precision = 1, Recall = 1, F1-score = 1. The identification accuracy for first-grade, second-grade and qualified salted sea cucumber were all 100%. This method provides a new idea for the grade evaluation of salted sea cucumber.

Key words: salted sea cucumber; vision and tactile sensation; dynamic energy map; mechanical characteristics; machine learning

CLC Number: