Hyperspectral imaging coupled with multivariate methods for seed vitality estimation and forecast for Quercus variabilis

An end-to-end seed vigor prediction model for imbalanced samples using hyperspectral image

Hyperspectral imaging is a key technology for non-destructive detection of seed vigor presently due to its capability to capture variations of optical properties in seeds. As the seed vigor data depends on the actual germination rate, it inevitably results in an imbalance between positive and negative samples. Additionally, hyperspectral image (HSI) suffers from feature redundancy and collinearity due to its inclusion of hundreds of wavelengths. It also creates a challenge to extract effective wavelength information in feature selection, however, which limits the ability of deep learning to extract features from HSI and accurately predict seed vigor. Accordingly, in this paper, we proposed a Focal-WAResNet network to predict seed vigor end-to-end, which improves the network performance and feature representation capability, and improves the accuracy of seed vigor prediction. Firstly, the focal loss function is utilized to adjust the loss weights of different sample categories to solve the problem of sample imbalance. Secondly, a WAResNet network is proposed to select characteristic wavelengths and predict seed vigor end-to-end, focusing on wavelengths with higher network weights, which enhance the ability of seed vigor prediction. To validate the effectiveness of this method, this study collected HSI of maize seeds for experimental verification, providing a reference for plant breeding. The experimental results demonstrate a significant improvement in classification performance compared to other state-of-the-art methods, with an accuracy up to 98.48% and an F1 score of 95.9%.

Non-destructive detection of single-seed viability in maize using hyperspectral imaging technology and multi-scale 3D convolutional neural network

The viability of Zea mays seed plays a critical role in determining the yield of corn. Therefore, developing a fast and non-destructive method is essential for rapid and large-scale seed viability detection and is of great significance for agriculture, breeding, and germplasm preservation. In this study, hyperspectral imaging (HSI) technology was used to obtain images and spectral information of maize seeds with different aging stages. To reduce data input and improve model detection speed while obtaining more stable prediction results, successive projections algorithm (SPA) was used to extract key wavelengths that characterize seed viability, then key wavelength images of maize seed were divided into small blocks with 5 pixels ×5 pixels and fed into a multi-scale 3D convolutional neural network (3DCNN) for further optimizing the discrimination possibility of single-seed viability. The final discriminant result of single-seed viability was determined by comprehensively evaluating the result of all small blocks belonging to the same seed with the voting algorithm. The results showed that the multi-scale 3DCNN model achieved an accuracy of 90.67% for the discrimination of single-seed viability on the test set. Furthermore, an effort to reduce labor and avoid the misclassification caused by human subjective factors, a YOLOv7 model and a Mask R-CNN model were constructed respectively for germination judgment and bud length detection in this study, the result showed that mean average precision (mAP) of YOLOv7 model could reach 99.7%, and the determination coefficient of Mask R-CNN model was 0.98. Overall, this study provided a feasible solution for detecting maize seed viability using HSI technology and multi-scale 3DCNN, which was crucial for large-scale screening of viable seeds. This study provided theoretical support for improving planting quality and crop yield.

Sugarbeet Seed Germination Prediction Using Hyperspectral Imaging Information Fusion

Germination rate is important for seed selection and planting and quality. In this study, hyperspectral image technology integrated with germination tests was applied for feature association analysis and germination performance prediction of sugarbeet seeds. In this study, we proposed a nondestructive prediction method for sugarbeet seed germination. Sugarbeet seed was studied, and hyperspectral imaging (HIS) performed by binarization, morphology, and contour extraction was applied as a nondestructive and accurate technique to achieve single seed image segmentation. Comparative analysis of nine spectral pretreatment methods, SNV + 1D was used to process the average spectrum of sugarbeet seeds. Fourteen characteristic wavelengths were obtained by the Kullback-Leibler (KL) divergence, as the spectral characteristics of sugarbeet seeds. Principal component analysis (PCA) and material properties verified the validity of the extracted characteristic wavelengths. It was extracted of six image features of the hyperspectral image of a single seed obtained based on the gray-level co-occurrence matrix (GLCM). The spectral features, image features, and fusion features were used to establish partial least squares discriminant analysis (PLS-DA), CatBoost, and support vector machine radial-basis function (SVM-RBF) models respectively to predict the germination. The results showed that the prediction effect of fusion features was better than spectral features and image features. By comparing other models, the prediction results of the CatBoost model accuracy were up to 93.52%. The results indicated that, based on HSI and fusion features, the prediction of germinating sugarbeet seeds was more accurate and nondestructive.

Hyperspectral Characteristic Band Selection and Estimation Content of Soil Petroleum Hydrocarbon Based on GARF-PLSR

With continuous improvements in oil production, the environmental problems caused by oil exploitation are becoming increasingly serious. Rapid and accurate estimation of soil petroleum hydrocarbon content is of great significance to the investigation and restoration of environments in oil-producing areas. In this study, the content of petroleum hydrocarbon and the hyperspectral data of soil samples collected from an oil-producing area were measured. For the hyperspectral data, spectral transforms, including continuum removal (CR), first- and second-order differential (CR-FD, CR-SD), and Napierian logarithm (CR-LN), were applied to eliminate background noise. At present, there are some shortcomings in the method of feature band selection, such as large quantity, time of calculation, and unclear importance of each feature band obtained. Meanwhile, redundant bands easily exist in the feature set, which seriously affects the accuracy of the inversion algorithm. In order to solve the above problems, a new method (GARF) for hyperspectral characteristic band selection was proposed. It combined the advantage that the grouping search algorithm can effectively reduce the calculation time with the advantage that the point-by-point search algorithm can determine the importance of each band, which provided a clearer direction for further spectroscopic research. The 17 selected bands were used as the input data of partial least squares regression (PLSR) and K-nearest neighbor (KNN) algorithms to estimate soil petroleum hydrocarbon content, and the leave-one-out method was used for cross-validation. The root mean squared error (RMSE) and coefficient of determination (R²) of the estimation result were 3.52 and 0.90, which implemented a high accuracy with only 8.37% of the entire bands. The results showed that compared with the traditional characteristic band selection methods, GARF can effectively reduce the redundant bands and screen out the optimal characteristic bands in the hyperspectral data of soil petroleum hydrocarbon with the method of importance assessment, which retained the physical meaning. It provided a new idea for the research of other substances in soil.

Identification of Defective Maize Seeds Using Hyperspectral Imaging Combined with Deep Learning

Seed quality affects crop yield and the quality of agricultural products, and traditional identification methods are time-consuming, complex, and irreversibly destructive. This study aims to establish a fast, non-destructive, and effective approach for defect detection in maize seeds based on hyperspectral imaging (HSI) technology combined with deep learning. Raw spectra collected from maize seeds (200 each healthy and worm-eaten) were pre-processed using detrending (DE) and multiple scattering correction (MSC) to highlight the spectral differences between samples. A convolutional neural network architecture (CNN-FES) based on a feature selection mechanism was proposed according to the importance of wavelength in the target classification task. The results show that the subset of 24 feature wavelengths selected by the proposed CNN-FES can capture important feature information in the spectral data more effectively than the conventional successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS) algorithms. In addition, a convolutional neural network architecture (CNN-ATM) based on an attentional classification mechanism was designed for one-dimensional spectral data classification and compared with three commonly used machine learning methods, linear discriminant analysis (LDA), random forest (RF), and support vector machine (SVM). The results show that the classification performance of the designed CNN-ATM on the full wavelength does not differ much from the above three methods, and the classification accuracy is above 90% on both the training and test sets. Meanwhile, the accuracy, sensitivity, and specificity of CNN-ATM based on feature wavelength modeling can reach up to 97.50%, 98.28%, and 96.77% at the highest, respectively. The study shows that hyperspectral imaging-based defect detection of maize seed is feasible and effective, and the proposed method has great potential for the processing and analysis of complex hyperspectral data.

Single Seed Identification in Three Medicago Species via Multispectral Imaging Combined with Stacking Ensemble Learning

Multispectral imaging (MSI) has become a new fast and non-destructive detection method in seed identification. Previous research has usually focused on single models in MSI data analysis, which always employed all features and increased the risk to efficiency and that of system cost. In this study, we developed a stacking ensemble learning (SEL) model for successfully identifying a single seed of sickle alfalfa (Medicago falcata), hybrid alfalfa (M. varia), and alfalfa (M. sativa). SEL adopted a three-layer structure, i.e., level 0 with principal component analysis (PCA), linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA) as models of dimensionality reduction and feature extraction (DRFE); level 1 with support vector machine (SVM), multiple logistic regression (MLR), generalized linear models with elastic net regularization (GLMNET), and eXtreme Gradient Boosting (XGBoost) as basic learners; and level 3 with XGBoost as meta-learner. We confirmed that the values of overall accuracy, kappa, precision, sensitivity, specificity, and sensitivity in the SEL model were all significantly higher than those in basic models alone, based on both spectral features and a combination of morphological and spectral features. Furthermore, we also developed a feature filtering process and successfully selected 5 optimal features out of 33 ones, which corresponded to the contents of chlorophyll, anthocyanin, fat, and moisture in seeds. Our SEL model in MSI data analysis provided a new way for seed identification, and the feature filter process potentially could be used widely for development of a low-cost and narrow-channel sensor.

Intelligent detection of hard seeds of snap bean based on hyperspectral imaging

As a common problem in snap beans, hard seed has seriously affected the large-scale industrial planting and yield of snap bean. To realize accurate, quick and non-destructive identifying the hard seeds of snap bean is of great significance to avoiding the effects of hard seeds on germination and growth. This research was based on hyperspectral imaging (HSI) to achieve accurate detection of hard seeds of snap bean. This study obtained the characteristic spectra from the hyperspectral image of a single seed, and then combined the synthetic minority over-sampling technique (SMOTE) and Tomek links to balance the numbers of hard and non-hard seed samples. The characteristic wavelengths were extracted from the average spectrum. Then the average spectrum was processed by first derivative (1D). After that, the characteristic wavelengths could be extracted using successive projections algorithm (SPA). Finally, a radial basis function-support vector machine (RBF-SVM) model was established to realize the intelligent detection of hard seeds, and the detection accuracy rate reached 89.32%. The research results showed that HSI technology could achieved accurate, fast and non-destructive testing of the hard seeds of snap bean, which is of great significance to the large-scale and standardized planting of snap bean and increase the yield per unit area.

Prediction of soil organic matter content based on characteristic band selection method

Soil organic matter (SOM) is a key index for evaluating soil fertility and plays a vital role in the terrestrial carbon cycle. Visible and near-infrared (Vis-NIR) spectroscopy is an effective method for determining soil properties and is often used to predict SOM content. However, the key prerequisite for effective prediction of SOM content by Vis-NIR spectroscopy lies in the selection of appropriate preprocessing methods and effective data mining techniques. Therefore, in this study, six commonly used spectral preprocessing methods and effective characteristic band selection methods were selected to process the spectrum to predict SOM content. This study aims to determine a stable spectral preprocessing method and explore the predictive performance of different characteristic band selection methods. The results showed that: (i) The first derivative (FD) is the most stable spectral preprocessing method that can effectively improve the spectral characteristic information and the prediction effect of the model. (ii) The prediction effect of SOM content based on characteristic band selection methods is generally better than the full-spectra data. (iii) The precision of FD preprocessing spectrum combined with successive projections algorithm (SPA) in the partial least square regression prediction model of SOM content is the best. (iv) Although the prediction effect of the model based on the optimal band combination algorithm is slightly lower than that of SPA, it shows stable prediction performance, which provides a feasible method for SOM content prediction. In summary, the characteristic band selection method combined with FD can significantly improve the prediction accuracy of SOM content.

Nondestructive determination of SSC in an apple by using a portable near-infrared spectroscopy system

The soluble solids content (SSC) is an important factor in the internal quality detection of apples. It is essential to have reliable and high-speed measurement of the SSC. However, almost all traditional equipment is inconvenient and expensive. We designed a handheld nondestructive SSC detector based on near-infrared (NIR) spectroscopy, which is composed of a portable NIR spectrometer, cloud server, smartphone app, and prediction model of SSC. We preprocessed the spectrum with multiplicative scatter correction (MSC), standard normal variable transformation (SNV), and Savitzky-Golay (S-G) smoothing algorithms. Besides, the linear weight reduction of the particle swarm optimization algorithm is carried out, and we establish the model of an extreme learning machine optimized with the improved particle swarm optimization (IPSO-ELM) algorithm. The R², root mean square error of prediction (RMSEP), and residual prediction deviation (RPD) of the model are 0.993, 0.0155, and 11.6, respectively, which are better than the traditional model obviously. In addition, the number of wavelengths reduced from 228 to 70 as the model is simplified with the uninformative variable elimination (UVE) method. The time of training is reduced by 29.30% compared with the original spectrum. It can be verified that the IPSO-ELM model has good prediction performance, and the NIR diffuse reflectance spectroscopy is a reliable nondestructive measurement of SSC in apples.

Non-Destructive Identification of Naturally Aged Alfalfa Seeds via Multispectral Imaging Analysis

Seed aging detection and viable seed prediction are of great significance in alfalfa seed production, but traditional methods are disposable and destructive. Therefore, the establishment of a rapid and non-destructive seed screening method is necessary in seed industry and research. In this study, we used multispectral imaging technology to collect morphological features and spectral traits of aging alfalfa seeds with different storage years. Then, we employed five multivariate analysis methods, i.e., principal component analysis (PCA), linear discrimination analysis (LDA), support vector machines (SVM), random forest (RF) and normalized canonical discriminant analysis (nCDA) to predict aged and viable seeds. The results revealed that the mean light reflectance was significantly different at 450~690 nm between non-aged and aged seeds. LDA model held high accuracy (99.8~100.0%) in distinguishing aged seeds from non-aged seeds, higher than those of SVM (87.4~99.3%) and RF (84.6~99.3%). Furthermore, dead seeds could be distinguished from the aged seeds, with accuracies of 69.7%, 72.0% and 97.6% in RF, SVM and LDA, respectively. The accuracy of nCDA in predicting the germination of aged seeds ranged from 75.0% to 100.0%. In summary, we described a nondestructive, rapid and high-throughput approach to screen aged seeds with various viabilities in alfalfa.