FOOD SCIENCE ›› 2026, Vol. 47 ›› Issue (11): 331-339.doi: 10.7506/spkx1002-6630-20251208-087

• Safety Detection • Previous Articles    

Geographical Origin Identification of Soybean Based on Feature Combination and Granular Neural Network

FU Xingyu, LIN Jian, WENG Peilin, GUO Jun, LIN Zhiwei, CHEN Zhiteng, CHEN Yiwei   

  1. (1. College of Computer and Information Sciences, Fujian Agriculture and Forestry University, Fuzhou 350002, China; 2. Digital Fujian Research Institute of Big Data for Agriculture and Forestry, Fujian Agriculture and Forestry University, Fuzhou 350002, China)
  • Published:2026-07-02

RichHTML

0

PDF (PC)

6

Abstract: This study proposed a method for identifying the geographical origin of soybean based on feature combination and a granular neural network, aiming to overcome traceability difficulties caused by single features and data uncertainty in soybeans from different production regions. In this study, 8 562 soybean samples from five different provinces were collected for extraction of their shape, color, and texture features. First, an improved information granulation method was used to granulate the data from these samples to construct granular vectors. Second, a granular activation function and a granular loss function were designed to improve the neural network structure. Compared with existing neural networks, this method could structurally process features of different granularities in parallel and output the optimal discriminant results through a competitive mechanism. Finally, a comparative experiment was conducted to evaluate the performance of six algorithms in identifying the geographical origin of soybean based on single features and feature combinations. The results indicated that the combination of shape, color, and texture features achieved the best discrimination effect. The granular neural network achieved an accuracy of 94.86% under this combination, which was 6.54%, 10.16%, and 5.8% higher than that using single shape, color, and texture features, respectively. Meanwhile, the performance of this granular neural network model was also superior to that of other mainstream algorithms, with an accuracy improvement of 2.96%, 3.62%, 2.53%, 2.33%, and 3% when compared with support vector machines, random forests, gradient boosting decision trees, extreme gradient boosting, and standard neural networks, respectively. The combined use of the feature combination method and the granular neural network model proposed in this study enables accurate identification of soybeans from different geographical origins, providing new solutions and technical ideas for the geographical origin traceability of agricultural products.

Key words: geographical origin traceability; granular computing; neural network; feature combination; soybean identification

CLC Number: