Discrimination of Fresh Tobacco Leaves with Different Maturity Levels by Near-Infrared (NIR) Spectroscopy and Deep Learning

In-Field Tobacco Leaf Maturity Detection with an Enhanced MobileNetV1: Incorporating a Feature Pyramid Network and Attention Mechanism

The maturity of tobacco leaves plays a decisive role in tobacco production, affecting the quality of the leaves and production control. Traditional recognition of tobacco leaf maturity primarily relies on manual observation and judgment, which is not only inefficient but also susceptible to subjective interference. Particularly in complex field environments, there is limited research on in situ field maturity recognition of tobacco leaves, making maturity recognition a significant challenge. In response to this problem, this study proposed a MobileNetV1 model combined with a Feature Pyramid Network (FPN) and attention mechanism for in situ field maturity recognition of tobacco leaves. By introducing the FPN structure, the model fully exploits multi-scale features and, in combination with Spatial Attention and SE attention mechanisms, further enhances the expression ability of feature map channel features. The experimental results show that this model, with a size of 13.7 M and FPS of 128.12, performed outstandingly well on the task of field maturity recognition of tobacco leaves, achieving an accuracy of 96.3%, superior to classical models such as VGG16, VGG19, ResNet50, and EfficientNetB0, while maintaining excellent computational efficiency and small memory footprint. Experiments were conducted involving noise perturbations, changes in environmental brightness, and occlusions to validate the model's robustness in dealing with the complex environments that may be encountered in actual applications. Finally, the Score-CAM algorithm was used for result visualization. Heatmaps showed that the vein and color variations of the leaves provide key feature information for maturity recognition. This indirectly validates the importance of leaf texture and color features in maturity recognition and, to some extent, enhances the credibility of the model. The model proposed in this study maintains high performance while having low storage requirements and computational complexity, making it significant for in situ field maturity recognition of tobacco leaves.

Overlapped tobacco shred image segmentation and area computation using an improved Mask RCNN network and COT algorithm

Introduction

The classification of the four tobacco shred varieties, tobacco silk, cut stem, expanded tobacco silk, and reconstituted tobacco shred, and the subsequent determination of tobacco shred components, are the primary tasks involved in calculating the tobacco shred blending ratio. The identification accuracy and subsequent component area calculation error directly affect the composition determination and quality of the tobacco shred. However, tiny tobacco shreds have complex physical and morphological characteristics; in particular, there is substantial similarity between the expanded tobacco silk and tobacco silk varieties, and this complicates their classification. There must be a certain amount of overlap and stacking in the distribution of tobacco shreds on the actual tobacco quality inspection line. There are 24 types of overlap alone, not to mention the stacking phenomenon. Self-winding does not make it easier to distinguish such varieties from the overlapped types, posing significant difficulties for machine vision-based tobacco shred classification and component area calculation tasks.

Methods

This study focuses on two significant challenges associated with identifying various types of overlapping tobacco shreds and acquiring overlapping regions to calculate overlapping areas. It develops a new segmentation model for tobacco shred images based on an improved Mask region-based convolutional neural network (RCNN). Mask RCNN is used as the segmentation network's mainframe. Convolutional network and feature pyramid network (FPN) in the backbone are replaced with Densenet121 and U-FPN, respectively. The size and aspect ratios of anchors parameters in region proposal network (RPN) are optimized. An algorithm for the area calculation of the overlapped tobacco shred region (COT) is also proposed, which is applied to overlapped tobacco shred mask images to obtain overlapped regions and calculate the overlapped area.

Results

The experimental results showed that the final segmentation accuracy and recall rates are 89.1% and 73.2%, respectively. The average area detection rate of 24 overlapped tobacco shred samples increases from 81.2% to 90%, achieving high segmentation accuracy and overlapped area calculation accuracy.

Discussion

This study provides a new implementation method for the type identification and component area calculation of overlapped tobacco shreds and a new approach for other similar overlapped image segmentation tasks.

Rapid Detection of Fraudulent Rice Using Low-Cost Digital Sensing Devices and Machine Learning

Rice fraud is one of the common threats to the rice industry. Conventional methods to detect rice adulteration are costly, time-consuming, and tedious. This study proposes the quantitative prediction of rice adulteration levels measured through the packaging using a handheld near-infrared (NIR) spectrometer and electronic nose (e-nose) sensors measuring directly on samples and paired with machine learning (ML) algorithms. For these purposes, the samples were prepared by mixing rice at different ratios from 0% to 100% with a 10% increment based on the rice's weight, consisting of (i) rice from different origins, (ii) premium with regular rice, (iii) aromatic with non-aromatic, and (iv) organic with non-organic rice. Multivariate data analysis was used to explore the sample distribution and its relationship with the e-nose sensors for parameter engineering before ML modeling. Artificial neural network (ANN) algorithms were used to predict the adulteration levels of the rice samples using the e-nose sensors and NIR absorbances readings as inputs. Results showed that both sensing devices could detect rice adulteration at different mixing ratios with high correlation coefficients through direct (e-nose; R = 0.94-0.98) and non-invasive measurement through the packaging (NIR; R = 0.95-0.98). The proposed method uses low-cost, rapid, and portable sensing devices coupled with ML that have shown to be reliable and accurate to increase the efficiency of rice fraud detection through the rice production chain.

Discrimination of Substandard and Falsified Formulations from Genuine Pharmaceuticals Using NIR Spectra and Machine Learning

Near-infrared (NIR) spectroscopy is a promising technique for field identification of substandard and falsified drugs because it is portable, rapid, nondestructive, and can differentiate many formulated pharmaceutical products. Portable NIR spectrometers rely heavily on chemometric analyses based on libraries of NIR spectra from authentic pharmaceutical samples. However, it is difficult to build comprehensive product libraries in many low- and middle-income countries due to the large numbers of manufacturers who supply these markets, frequent unreported changes in materials sourcing and product formulation by the manufacturers, and general lack of cooperation in providing authentic samples. In this work, we show that a simple library of lab-formulated binary mixtures of an active pharmaceutical ingredient (API) with two diluents gave good performance on field screening tasks, such as discriminating substandard and falsified formulations of the API. Six data analysis models, including principal component analysis and support-vector machine classification and regression methods and convolutional neural networks, were trained on binary mixtures of acetaminophen with either lactose or ascorbic acid. While the models all performed strongly in cross-validation (on formulations similar to their training set), they individually showed poor robustness for formulations outside the training set. However, a predictive algorithm based on the six models, trained only on binary samples, accurately predicts whether the correct amount of acetaminophen is present in ternary mixtures, genuine acetaminophen formulations, adulterated acetaminophen formulations, and falsified formulations containing substitute APIs. This data analytics approach may extend the utility of NIR spectrometers for analysis of pharmaceuticals in low-resource settings.

Rapid determination of the total content of oleanolic acid and ursolic acid in Chaenomelis Fructus using near-infrared spectroscopy

Chaenomelis Fructus is a widely used traditional Chinese medicine with a long history in China. The total content of oleanolic acid (OA) and ursolic acid (UA) is taken as an important quality marker of Chaenomelis Fructus. In this study, quantitative models for the prediction total content of OA and UA in Chaenomelis Fructus were explored based on near-infrared spectroscopy (NIRS). The content of OA and UA in each sample was determined using high-performance liquid chromatography (HPLC), and the data was used as a reference. In the partial least squares (PLS) model, both leave one out cross validation (LOOCV) of the calibration set and external validation of the validation set were used to screen spectrum preprocessing methods, and finally the multiplicative scatter correction (MSC) was chosen as the optimal pretreatment method. The modeling spectrum bands and ranks were optimized using PLS regression, and the characteristic spectrum range was determined as 7,500-4,250 cm^-1, with 14 optimal ranks. In the back propagation artificial neural network (BP-ANN) model, the scoring data of 14 ranks obtained from PLS regression analysis were taken as input variables, and the total content of OA and UA reference values were taken as output values. The number of hidden layer nodes of BP-ANN was screened by full-cross validation (Full-CV) of the calibration set and external validation of the validation set. The result shows that both PLS model and PLS-BP-ANN model have strong prediction ability. In order to evaluate and compare the performance and prediction ability of models, the total content of OA and UA in each sample of the test set were detected under the same HPLC conditions, the NIRS data of the test set were input, respectively, to the optimized PLS model and PLS-BP-ANN model. By comparing the root-mean-square error (RMSEP) and determination coefficient (R ²) of the test set and ratio of performance to deviation (RPD), the PLS-BP-ANN model was found to have better performance with RMSEP of 0.59 mg·g^-1, R ² of 95.10%, RPD of 4.53 and bias of 0.0387 mg·g^-1. The results indicated that NIRS can be used for the rapid quality control of Chaenomelis Fructus.

Rapid monitoring of flavonoid content in sweet tea (Lithocarpus litseifolius (Hance) Chun) leaves using NIR spectroscopy

BACKGROUND

Sweet tea, which functions as tea, sugar and medicine, was listed as a new food resource in 2017. Flavonoids are the main medicinal components in sweet tea and have significant pharmacological activities. Therefore, the quality of sweet tea is related to the content of flavonoids. Flavonoid content in plants is normally determined by time-consuming and expensive chemical analyses. The aim of this study was to develop a methodology to measure three constituents of flavonoids, namely, total flavonoids, phloridin and trilobatin, in sweet tea leaves using near-infrared spectroscopy (NIR).

RESULTS

In this study, we demonstrated that the combination of principal component analysis (PCA) and NIR spectroscopy can distinguish sweet tea from different locations. In addition, different spectral preprocessing methods are used to establish partial least squares (PLS) models between spectral information and the content of the three constituents. The best total flavonoid prediction model was obtained with NIR spectra preprocessed with Savitzky-Golay combined with second derivatives (SG + D2) (R_P² = 0.893, and RMSEP = 0.131). For trilobatin, the model with the best performance was developed with raw NIR spectra (R_P² = 0.902, and RMSEP = 2.993), and for phloridin, the best model was obtained with NIR spectra preprocessed with standard normal variate (SNV) (R_P² = 0.818, and RMSEP = 1.085). The coefficients of determination for all calibration sets, validation sets and prediction sets of the best PLS models were higher than 0.967, 0.858 and 0.818, respectively.

CONCLUSIONS

The conclusion indicated that NIR spectroscopy has the ability to determine the flavonoid content of sweet tea quickly and conveniently.

Identification of Amaranthus Species Using Visible-Near-Infrared (Vis-NIR) Spectroscopy and Machine Learning Methods

The feasibility of rapid and non-destructive classification of six different Amaranthus species was investigated using visible-near-infrared (Vis-NIR) spectra coupled with chemometric approaches. The focus of this research would be to use a handheld spectrometer in the field to classify six Amaranthus sp. in different geographical regions of South Korea. Spectra were obtained from the adaxial side of the leaves at 1.5 nm intervals in the Vis-NIR spectral range between 400 and 1075 nm. The obtained spectra were assessed with four different preprocessing methods in order to detect the optimum preprocessing method with high classification accuracy. Preprocessed spectra of six Amaranthus sp. were used as input for the machine learning-based chemometric analysis. All the classification results were validated using cross-validation to produce robust estimates of classification accuracies. The different combinations of preprocessing and modeling were shown to have a classification accuracy of between 71% and 99.7% after the cross-validation. The combination of Savitzky-Golay preprocessing and Support vector machine showed a maximum mean classification accuracy of 99.7% for the discrimination of Amaranthus sp. Considering the high number of spectra involved in this study, the growth stage of the plants, varying measurement locations, and the scanning position of leaves on the plant are all important. We conclude that Vis-NIR spectroscopy, in combination with appropriate preprocessing and machine learning methods, may be used in the field to effectively classify Amaranthus sp. for the effective management of the weedy species and/or for monitoring their food applications.