A method of information fusion for identification of rice seed varieties based on hyperspectral imaging technology

Abstract

Accurate, rapid, and nondestructive identification of rice seed varieties has great significance for agriculture and food security, a method based on information fusion and artificial fish swarm algorithm (AFSA) combined with the hyperspectral imaging (HSI) of five kinds of rice seeds was proposed in this work. First, the spectral and image data were obtained from HSI, and the spectral data were preprocessed by detrending. Then, bootstrapping soft shrinkage (BOSS), variable iterative space shrinkage approach, successive projections algorithm, and principal component analysis were adopted to select feature variables from the spectral and image data. Next, the support vector machine (SVM) model was constructed based on the spectral and image feature variables. In order to further improve the classification accuracy of single feature model, the model based on fused feature was developed and it was finally optimized by AFSA. The research showed that the feature variables (114 spectral variables and 11 image variables) selected by BOSS were representative, and the model accuracy based on BOSS spectral and image features reached 91.48% and 70%, respectively. The performance of the SVM model based on fused feature was improved significantly, and the model accuracy reached 97.22%. After AFSA optimization, the model accuracy finally reached 99.44%. The above result confirmed that using AFSA to optimize the model based on fused feature could be a promising method to identify rice seed varieties. Practical application Rapid and accurate identification of rice seed varieties contributes to establishment of online rice seed identification system. HSI combined with information fusion and AFSA can overcome the disadvantage of low accuracy of traditional nondestructive testing methods. The method proposed in this article can be recommended to be widely popularized in farms, grain markets, and market regulators.

Hyperspectral imaging with machine learning for non-destructive classification of Astragalus membranaceus var. mongholicus, Astragalus membranaceus, and similar seeds

“…However, the reflectance of AMM was generally higher than that of AM seeds, with greater differences in the NIR region (780-1000 nm) than in the VIS region (400-780 nm). The VIS region may be related to β-carotene and anthocyanin in the seeds ( Sun et al., 2021 ), and the difference in the NIR region may be related to protein, starch and other organic matter in the seeds ( Cen and He, 2007 ; Awanthi et al., 2019 ). These results indicated that HSI could capture differences in texture, pigment and other physical and chemical properties between AMM and AM seeds.…”

“…Machine learning methods have been developed that are currently the most efficient approaches for image processing and analysis. Common machine learning algorithms, including support vector machine (SVM), partial least squares discriminant analysis (PLS-DA), and multilayer perceptron (MLP), have been successfully applied to a range of classification tasks ( Yang et al., 2015a ; Sun et al., 2016 ; Sun et al., 2021 ; Nazari et al., 2021 ).…”

Nondestructive detection of adulterated wolfberry (Lycium Chinense) fruits based on hyperspectral imaging technology

“…HSI technology has been explored in the food industry as a stopgap measure in detecting quality crops and related agriproducts (Khan et al, 2021; Nirere et al, 2021; J. Sun et al, 2021). Tang et al (2021) obtained 88.33% prediction accuracy from HSIs of six varieties of L. barbarum grains using competitive adaptive reweighted sampling (CARS) algorithms for wavelength selection, the whale optimization algorithm for model enhancement, and the SVM as the prediction model.…”

“…Notwithstanding HSI's effectiveness in nondestructive detection of food safety and quality, the workload involved in the spectra processing is relatively high, due to the various series of images that contain large amounts of data, therefore increasing complexity and causing poor model performance (Nirere et al, 2022; J. Sun et al, 2021). Hence, multivariate selection approaches, such as CARS, linear discriminant analysis (LDA), variable iterative space shrinkage approach, and so forth, are frequently utilized to reduce the amount of data for improved model performance.…”

Identification of maize seed varieties based on stacked sparse autoencoder and near‐infrared hyperspectral imaging technology

“…Different seed varieties can be identified by analyzing the changes in the NIR spectra of different seed varieties (Zhou et al, 2020). In the previous literature, the NIR hyperspectral imaging technology was employed for seed vigor detection (Al‐Amery et al, 2018), seed moisture content detection (Ferreira et al, 2014; Wang et al, 2020), and seed varieties classification (Sun et al, 2021; Xie & He, 2018). The results suggested that NIR hyperspectral imaging technology was an effective method for seed variety identification.…”