Variety classification of coated maize seeds based on Raman hyperspectral imaging

ATR-FTIR spectroscopy combined with chemometrics to assess the spectral markers of irradiated baijius and their potential application in irradiation dose control

Although some methods have been proposed for the identification of irradiated baijius, they are often costly, time-consuming, and destructive. It is also unclear what instrumentation can be used to fully characterize the quality changes in irradiated baijius. To address this issue, this study pioneers the use of attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy in combination with chemometrics to open up new avenues for characterizing irradiated baijius and their quality control. Principal component analysis, five spectral pre-processing methods (Savitzky-Golay smoothing (S-G), second-order derivative (SD), multiple scattering correction (MSC), S-G + SD and S-G + MSC), five wavelength selection methods (random forest variable importance (RFVI), two-dimensional correlation spectroscopy (2D-COS), variable importance in projection (VIP), ReliefF, and Venn), and three classification models (partial least squares-discriminant analysis (PLS-DA), random forest (RF), and grasshopper optimization algorithm-based support vector machine (GOA-SVM)) were integrated into the analytical framework of ATR-FTIR spectroscopy, aiming to accurately identify baijiu samples according to different irradiation doses and to search for irradiation-induced spectral difference characteristics (spectral markers). The results showed that SD was the best spectral pre-processing method, and RF models constructed using the 20 most competitive and discriminative spectral markers (selected by Venn) could achieve accurate identification of baijiu samples based on irradiation dose (0, 4, 6, and 8 kGy). After Pearson correlation analysis, the five significantly (P＜0.05) changed spectral markers (1596, 2025, 2309, 2329, and 2380 cm-1) were attributed to changes in the content of total acids, alcohols, and aromatic compounds. These findings demonstrate for the first time the potential of ATR-FTIR spectroscopy as a fast, low-cost, and non-destructive tool for the characterization and identification of irradiated baijiu samples. This approach may also offer a promising solution for labeling management of irradiated foods, vintage identification of baijius, and brand protection.

ADFSNet: An Adaptive Domain Feature Separation Network for the Classification of Wheat Seed Using Hyperspectral Images

Wheat seed classification is a critical task for ensuring crop quality and yield. However, the characteristics of wheat seeds can vary due to variations in climate, soil, and other environmental factors across different years. Consequently, the present classification model is no longer adequate for accurately classifying novel samples. To tackle this issue, this paper proposes an adaptive domain feature separation (ADFS) network that utilizes hyperspectral imaging techniques for cross-year classification of wheat seed varieties. The primary objective is to improve the generalization ability of the model at a minimum cost. ADFS leverages deep learning techniques to acquire domain-irrelevant features from hyperspectral data, thus effectively addressing the issue of domain shifts across datasets. The feature spaces are divided into three parts using different modules. One shared module aligns feature distributions between the source and target datasets from different years, thereby enhancing the model's generalization and robustness. Additionally, two private modules extract class-specific features and domain-specific features. The transfer mechanism does not learn domain-specific features to reduce negative transfer and improve classification accuracy. Extensive experiments conducted on a two-year dataset comprising four wheat seed varieties demonstrate the effectiveness of ADFS in wheat seed classification. Compared with three typical transfer learning networks, ADFS can achieve the best accuracy of wheat seed classification with small batch samples updated, thereby addressing new seasonal variability.

Protein and lipid content estimation in soybeans using Raman hyperspectral imaging

Unlike standard chemical analysis methods involving time-consuming, labor-intensive, and invasive pretreatment procedures, Raman hyperspectral imaging (HSI) can rapidly and non-destructively detect components without professional supervision. Generally, the Kjeldahl methods and Soxhlet extraction are used to chemically determine the protein and lipid content of soybeans. This study is aimed at developing a high-performance model for estimating soybean protein and lipid content using a non-destructive Raman HSI. Partial least squares regression (PLSR) techniques were used to develop the model using a calibration model based on 70% spectral data, and the remaining 30% of the data were used for validation. The results indicate that the Raman HSI, combined with PLSR, resulted in a protein and lipid model Rp2 of 0.90 and 0.82 with Root Mean Squared Error Prediction (RMSEP) 1.27 and 0.79, respectively. Additionally, this study successfully used the Raman HSI approach to create a prediction image showing the distribution of the targeted components, and could predict protein and lipid based on a single seeds.

Self-supervised maize kernel classification and segmentation for embryo identification

Introduction

Computer vision and deep learning (DL) techniques have succeeded in a wide range of diverse fields. Recently, these techniques have been successfully deployed in plant science applications to address food security, productivity, and environmental sustainability problems for a growing global population. However, training these DL models often necessitates the large-scale manual annotation of data which frequently becomes a tedious and time-and-resource- intensive process. Recent advances in self-supervised learning (SSL) methods have proven instrumental in overcoming these obstacles, using purely unlabeled datasets to pre-train DL models.

Methods

Here, we implement the popular self-supervised contrastive learning methods of NNCLR Nearest neighbor Contrastive Learning of visual Representations) and SimCLR (Simple framework for Contrastive Learning of visual Representations) for the classification of spatial orientation and segmentation of embryos of maize kernels. Maize kernels are imaged using a commercial high-throughput imaging system. This image data is often used in multiple downstream applications across both production and breeding applications, for instance, sorting for oil content based on segmenting and quantifying the scutellum's size and for classifying haploid and diploid kernels.

Results and discussion

We show that in both classification and segmentation problems, SSL techniques outperform their purely supervised transfer learning-based counterparts and are significantly more annotation efficient. Additionally, we show that a single SSL pre-trained model can be efficiently finetuned for both classification and segmentation, indicating good transferability across multiple downstream applications. Segmentation models with SSL-pretrained backbones produce DICE similarity coefficients of 0.81, higher than the 0.78 and 0.73 of those with ImageNet-pretrained and randomly initialized backbones, respectively. We observe that finetuning classification and segmentation models on as little as 1% annotation produces competitive results. These results show SSL provides a meaningful step forward in data efficiency with agricultural deep learning and computer vision.

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

The roots of Astragalus membranaceus var. mongholicus (AMM) and A. membranaceus (AM) are widely used in traditional Chinese medicine. Although AMM has higher yields and accounts for a larger market share, its cultivation is fraught with challenges, including mixed germplasm resources and widespread adulteration of commercial seeds. Current methods for distinguishing Astragalus seeds from similar (SM) seeds are time-consuming, laborious, and destructive. To establish a non-destructive method, AMM, AM, and SM seeds were collected from various production areas. Machine vision and hyperspectral imaging (HSI) were used to collect morphological data and spectral data of each seed batch, which was used to establish discriminant models through various algorithms. Several preprocessing methods based on hyperspectral data were compared, including multiplicative scatter correction (MSC), standard normal variable (SNV), and first derivative (FD). Then selection methods for identifying informative features in the above data were compared, including successive projections algorithm (SPA), uninformative variable elimination (UVE), and competitive adaptive reweighted sampling (CARS). The results showed that support vector machine (SVM) modeling of machine vision data could distinguish Astragalus seeds from SM with >99% accuracy, but could not satisfactorily distinguish AMM seeds from AM. The FD-UVE-SVM model based on hyperspectral data reached 100.0% accuracy in the validation set. Another 90 seeds were tested, and the recognition accuracy was 100.0%, supporting the stability of the model. In summary, HSI data can be applied to discriminate among the seeds of AMM, AM, and SM non-destructively and with high accuracy, which can drive standardization in the Astragalus production industry.

Rapid nondestructive detection of peanut varieties and peanut mildew based on hyperspectral imaging and stacked machine learning models

Moldy peanut seeds are damaged by mold, which seriously affects the germination rate of peanut seeds. At the same time, the quality and variety purity of peanut seeds profoundly affect the final yield of peanuts and the economic benefits of farmers. In this study, hyperspectral imaging technology was used to achieve variety classification and mold detection of peanut seeds. In addition, this paper proposed to use median filtering (MF) to preprocess hyperspectral data, use four variable selection methods to obtain characteristic wavelengths, and ensemble learning models (SEL) as a stable classification model. This paper compared the model performance of SEL and extreme gradient boosting algorithm (XGBoost), light gradient boosting algorithm (LightGBM), and type boosting algorithm (CatBoost). The results showed that the MF-LightGBM-SEL model based on hyperspectral data achieves the best performance. Its prediction accuracy on the data training and data testing reach 98.63% and 98.03%, respectively, and the modeling time was only 0.37s, which proved that the potential of the model to be used in practice. The approach of SEL combined with hyperspectral imaging techniques facilitates the development of a real-time detection system. It could perform fast and non-destructive high-precision classification of peanut seed varieties and moldy peanuts, which was of great significance for improving crop yields.

Detection of seed purity of hybrid wheat using reflectance and transmittance hyperspectral imaging technology

Chemical hybridization and genic male sterility systems are two main methods of hybrid wheat production; however, complete sterility of female wheat plants cannot be guaranteed owing to the influence of the growth stage and weather. Consequently, hybrid wheat seeds are inevitably mixed with few parent seeds, especially female seeds. Therefore, seed purity is a key factor in the popularization of hybrid wheat. However, traditional seed purity detection and variety identification methods are time-consuming, laborious, and destructive. Therefore, to establish a non-destructive classification method for hybrid and female parent seeds, three hybrid wheat varieties (Jingmai 9, Jingmai 11, and Jingmai 183) and their parent seeds were sampled. The transmittance and reflectance spectra of all seeds were collected via hyperspectral imaging technology, and a classification model was established using partial least squares-discriminant analysis (PLS-DA) combined with various preprocessing methods. The transmittance spectrum significantly improved the classification of hybrids and female parents compared to that obtained using reflectance spectrum. Specifically, using transmittance spectrum combined with a characteristic wavelength-screening algorithm, the Detrend-CARS-PLS-DA model was established, and the accuracy rates in the testing sets of Jingmai 9, Jingmai 11, and Jingmai 183 were 95.69%, 98.25%, and 97.25%, respectively. In conclusion, transmittance hyperspectral imaging combined with a machine learning algorithm can effectively distinguish female parent seeds from hybrid seeds. These results provide a reference for rapid seed purity detection in the hybrid production process. Owing to the non-destructive and rapid nature of hyperspectral imaging, the detection of hybrid wheat seed purity can be improved by online sorting in the future.

Nondestructive Classification of Maize Moldy Seeds by Hyperspectral Imaging and Optimal Machine Learning Algorithms

Mildew of maize seeds may affect their germination rates and reduce crop quality. It is crucial to classify maize seeds efficiently and without destroying their original structure. This study aimed to establish hyperspectral datasets using hyperspectral imaging (HSI) of maize seeds with different degrees of mildew and then classify them using spectral characteristics and machine learning algorithms. Initially, the images were processed with Otus and morphological operations. Each seed's spectral features were extracted based on its coding, its edge, region of interest (ROI), and original pixel coding. Random forest (RF) models were optimized using the sparrow search algorithm (SSA), which is incapable of escaping the local optimum; hence, it was optimized using a modified reverse sparrow search algorithm (JYSSA) strategy. This reverse strategy selects the top 10% as the elite group, allowing us to escape from local optima while simultaneously expanding the range of the sparrow search algorithm's optimal solution. Finally, the JYSSA-RF algorithm was applied to the validation set, with 96% classification accuracy, 100% precision, and a 93% recall rate. This study provides novel ideas for future nondestructive detection of seeds and moldy seed selection by combining hyperspectral imaging and JYSSA algorithms based on optimized RF.

Rapid and Non-destructive Classification of New and Aged Maize Seeds Using Hyperspectral Image and Chemometric Methods

The aged seeds have a significant influence on seed vigor and corn growth. Therefore, it is vital for the planting industry to identify aged seeds. In this study, hyperspectral reflectance imaging (1,000-2,000 nm) was employed for identifying aged maize seeds using seeds harvested in different years. The average spectra of the embryo side, endosperm side, and both sides were extracted. The support vector machine (SVM) algorithm was used to develop classification models based on full spectra to evaluate the potential of hyperspectral imaging for maize seed detection and using the principal component analysis (PCA) and ANOVA to reduce data dimensionality and extract feature wavelengths. The classification models achieved perfect performance using full spectra with an accuracy of 100% for the prediction set. The performance of models established with the first three principal components was similar to full spectrum models, but that of PCA loading models was worse. Compared to other spectra, the two-band ratio (1,987 nm/1,079 nm) selected by ANOVA from embryo-side spectra achieved a better classification accuracy of 95% for the prediction set. The image texture features, including histogram statistics (HS) and gray-level co-occurrence matrix (GLCM), were extracted from the two-band ratio image to establish fusion models. The results demonstrated that the two-band ratio selected from embryo-side spectra combined with image texture features achieved the classification of maize seeds harvested in different years with an accuracy of 97.5% for the prediction set. The overall results indicated that combining the two wavelengths with image texture features could detect aged maize seeds effectively. The proposed method was conducive to the development of multi-spectral detection equipment.