Mixed pesticide recognition based on three-dimensional fluorescence spectroscopy and a convolutional neural network.

Kasugamycin, spinosad, and lambda-cyhalothrin are common organic pesticides that are widely used to control and prevent diseases and pests in fruits and vegetables. However, the unreasonable use of pesticides will cause great harm to the natural environment and human health. Pesticides often exist in the form of mixtures in nature. Establishing recognition models for mixed pesticides in large-scale sample testing can provide guidance for further precise analysis and reduce resource waste and time. Therefore, finding a fast and effective identification method for mixed pesticides is of great significance. This paper applies three-dimensional fluorescence spectroscopy to detect mixed pesticides and introduces a convolutional neural network (CNN) model structure based on an improved LeNet-5 to classify mixed pesticides. The input part of the model corresponds to fluorescence spectrum data at excitation wavelengths of 250-306 nm and emission wavelengths of 300-450 nm, and the mixed pesticides are divided into three categories. The research results show that when the learning rate is set to 1 and the number of iterations is 300, the CNN classification model has ideal performance (with a recognition accuracy of 100%) and is superior to the performance of the support vector machine method. This paper provides a certain methodological basis for the rapid identification of mixed pesticides.

Self-denoising method for OCT images with single spectrogram-based deep learning.

The presence of noise in images reconstructed with optical coherence tomography (OCT) is a key issue which limits the further improvement of the image quality. In this Letter, for the first time, to the best of our knowledge, a self-denoising method for OCT images is presented with single spectrogram-based deep learning. Different noises in different images could be customized with an extremely low computation. The deep-learning model consists of two fully connected layers, two convolution layers, and one deconvolution layer, with the input being the raw interference spectrogram and the label being its reconstructed image using the Fourier transform. The denoising image could be calculated by subtracting the noise predicted by our model from the label image. The OCT images of the TiO2 phantom, the orange, and the zebrafish obtained with our spectral-domain OCT system are used as examples to demonstrate the capability of our method. The results demonstrate its effectiveness in reducing noises such as speckle patterns and horizontal and vertical stripes. Compared with the label image, the signal-to-noise ratio could be improved by 35.0 dB, and the image contrast could be improved by a factor of two. Compared with the results denoised by the average method, the mean peak signal-to-noise ratio is 26.2 dB.

GPU-accelerated image registration algorithm in ophthalmic optical coherence tomography.

Limited to the power of the light source in ophthalmic optical coherence tomography (OCT), the signal-to-noise ratio (SNR) of the reconstructed images is usually lower than OCT used in other fields. As a result, improvement of the SNR is required. The traditional method is averaging several images at the same lateral position. However, the image registration average costs too much time, which limits its real-time imaging application. In response to this problem, graphics processing unit (GPU)-side kernel functions are applied to accelerate the reconstruction of the OCT signals in this paper. The SNR of the images reconstructed from different numbers of A-scans and B-scans were compared. The results demonstrated that: 1) There is no need to realize the axial registration with every A-scan. The number of the A-scans used to realize axial registration is suitable to set as ∼25, when the A-line speed was set as ∼12.5kHz. 2) On the basis of ensuring the quality of the reconstructed images, the GPU can achieve 43× speedup compared with CPU.

Detection of Ningnanmycin Using Fluorescence Spectroscopy Combined with BP Neural Network.

BACKGROUND: Ningnanmycin is a new antibiotic pesticide with good bactericidal and antiviral efficacy, which is widely used in the control of fruit and vegetable diseases, and the excessive pesticide residues pose a serious threat to the environment and human health. METHODS: In this study, we used fluorescence spectrometer to scan the three-dimensional spectrum of ningnanmycin samples. We used a BP neural network to complete the regression analysis of content prediction based on the fluorescence spectra. After that, the prediction performance of the BP neural network was compared with the exponential fitting method. RESULTS: The results of the BP neural network modeling based on the obtained samples showed that the mean square error of the prediction results of the test set is less than 10-4, the R-square is greater than 0.99, the average recovery is 99.11%, and the model performance of the BP neural network is better than exponential fitting. CONCLUSION: Studies have shown that fluorescence spectroscopy combined with BP neural network can effectively predict the concentration of ningnanmycin.

Prediction of the Nitrogen Content of Rice Leaf Using Multi-Spectral Images Based on Hybrid Radial Basis Function Neural Network and Partial Least-Squares Regression.

This paper's novel focus is predicting the leaf nitrogen content of rice during growing and maturing. A multispectral image processing-based prediction model of the Radial Basis Function Neural Network (RBFNN) model was proposed. Moreover, this paper depicted three primary points as the following: First, collect images of rice leaves (RL) from a controlled condition experimental laboratory and new shoot leaves in different stages in the visible light spectrum, and apply digital image processing technology to extract the color characteristics of RL and the morphological characteristics of the new shoot leaves. Secondly, the RBFNN model, the General Regression Model (GRL), and the General Regression Method (GRM) model were constructed based on the extracted image feature parameters and the nitrogen content of rice leaves. Third, the RBFNN is optimized by and Partial Least-Squares Regression (RBFNN-PLSR) model. Finally, the validation results show that the nitrogen content prediction models at growing and mature stages that the mean absolute error (MAE), the Mean Absolute Percentage Error (MAPE), and the Root Mean Square Error (RMSE) of the RFBNN model during the rice-growing stage and the mature stage are 0.6418 (%), 0.5399 (%), 0.0652 (%), and 0.3540 (%), 0.1566 (%), 0.0214 (%) respectively, the predicted value of the model fits well with the actual value. Finally, the model may be used to give the best foundation for achieving exact fertilization control by continuously monitoring the nitrogen nutrition status of rice. In addition, at the growing stage, the RBFNN model shows better results compared to both GRL and GRM, in which MAE is reduced by 0.2233% and 0.2785%, respectively.

Detection of Dihydrocoumarin in Coconut Juice via Photoelectric Detection System Based on Ultraviolet Absorption Spectrometry.

Based on ultraviolet absorption spectroscopy technology combined with stoichiometry, a portable photoelectric detection system with wireless transmission was designed with the advantages of simple operation, low cost, and quick response to realize the non-destructive detection of dihydrocoumarin content in coconut juice. Through the detection of a sample solution, the light intensity through the solution is measured and converted into absorbance. Particle swarm optimization (PSO) is applied to optimize support vector regression (SVR) to establish a corresponding concentration prediction model. At the same time, in order to solve the shortcomings of the conventional portable photoelectric detection equipment in data storage, data transmission, and other aspects, based on the optimal PSO-SVR model, we used Python language to develop a friendly graphical user interface (GUI), integrating data collection, storage, analysis, and prediction modeling in one, greatly simplifying the operation process. The experimental results show that, compared with the traditional methods, the system achieves the purpose of rapid and non-destructive detection and has a small gap compared with the detection results of the ultraviolet spectrophotometer. It provides a good method for the determination of dihydrocoumarin in coconut juice.

Non-contact fluorescent detection of pesticide residues based on segment prediction using PLS and a curve fitting algorithm.

Fluorescence spectral analysis is an important method to detect the pesticide residues, which is vital for food safety issues. It has been demonstrated that the traditional curve fitting (CF) method can predict the concentration of pesticide with a high accuracy. However, low absorption of the samples at low concentration of pesticide is required; moreover, the pre-process of fruit juice is time-consuming and destructive to the samples. To overcome these disadvantages while maintaining the high accuracy in the high concentration range, the segment detection method is proposed in this paper. Two models were employed to predict the concentration according to the fluorescence intensity. The partial least squares (PLS) model was used to predict the concentration of the samples when the fluorescence intensity at 356 nm was smaller than 1, while the CF method was used to predict the concentration of samples when the fluorescence intensity at 356 nm was larger than 1 in our system. In total, 101 samples with concentration ranging from 0 to 0.0714 mg/mL were used to validate this method. The results indicated that the PLS method exhibited a high sensitivity in the low concentration range, while the CF method exhibited high accuracy in the high concentration range.

Detection of captan residues in apple juice using fluorescence spectroscopy combined with a genetic algorithm and support vector machines.

The captan residues in apple juice were detected by fluorescence spectrometry, and the content level of captan was predicted based on a genetic algorithm and support vector machines (GA-SVMs). According to the captan concentration in apple juice, the experimental samples were divided into four levels, including no excess, slight excess, moderate excess, and severe excess. A GA was used to select the characteristic wavelength and optimize SVM parameters, and SVM was applied to train the classification model. 50 characteristic wavelength points were selected from the original fluorescence spectra, which contained 401 wavelength points, and the classification accuracy of the training set and test set is 99.02% and 100%, respectively, which is higher than the traditional PLS method. The results show that a GA can effectively select the feature wavelengths, and an SVM model can accurately predict the content level of captan residues. A fast and non-destructive analysis method, combined with a GA and SVM based on fluorescence spectroscopy, was realized.

Analysis of pesticide residues by a support vector machine combined with fluorescence spectroscopy.

Pesticide residues enter a lake through the water cycle, causing harm to the water environment and human health. It is necessary to select highly sensitive fluorescence spectroscopy to detect pesticides (bifenthrin, prochloraz, and cyromazine), and a support vector machine (SVM) is used to analyze the concentration of pesticides. In addition, this paper adopts K-fold cross validation and a grid search to optimize the SVM algorithm. The performance evaluation index and running time prove the reliability of the results of this experiment. They show that fluorescence spectroscopy combined with SVM is efficient in predicting pesticide residue content.

Echo decomposition of full-waveform LiDAR based on a digital implicit model and a particle swarm optimization.

In the waveform decomposition of full-waveform LiDAR, the Gaussian model (GSM) and the generalized Gaussian model (GGSM) are widely used to extract echoes from return waveforms. However, those models have explicit functions that follow specific distribution shapes, so they are suitable only for decomposing echo waveforms with similar shapes. This paper introduces a digital implicit model (DIM) and presents a universal decomposition method for the full-waveform LiDAR. In this method, the decomposition model is considered to be an implicit function, associated with a digital template waveform library, whose optimization is implemented by a modified particle swarm algorithm. The template waveform library is a customized fingerprint for any special full-waveform LiDAR, so the DIM can deal effectively with infinite echoes with arbitrary shapes. A full-waveform LiDAR system with asymmetric echo distribution is designed to compare the decomposition performances among the GSM, GGSM, and DIM. Experimental results show that, when decomposing the return waveform containing a single echo, the normalized sum of squares due to fitting error (SSE) of the DIM can be 60 times lower than the GSM and the GGSM. By comparing the estimation accuracies of the amplitude, the FWHM and the location of the echo component, the DIM has the best decomposition performance and the best ranging accuracy (subcentimeter level) among the three models; when decomposing the return waveform containing three overlapping echoes, the normalized SSE of the DIM can be 28 times lower than the GSM and 12 times lower than the GGSM. By comparing the estimation accuracies of the amplitude, FWHM, and location of echoes components, the DIM has the best decomposition performance and best ranging accuracy (centimeter level) among the three models.

Multiple kinds of pesticide residue detection using fluorescence spectroscopy combined with partial least-squares models.

Compared with high-performance liquid chromatography and mass spectroscopy, fluorescence spectroscopy has attracted considerable attention in the field of pesticide residue detection due to its practical advantages of providing rapid, simultaneous analysis and non-destructive detection. However, given that the concentration of pesticide residue detected via fluorescence spectroscopy is calculated in accordance with the Beer-Lambert law, this method can only detect samples containing a single kind of pesticide or several kinds of pesticides with completely different fluorescences. Multiple partial least-squares (PLS) models are introduced in this work to overcome this disadvantage and achieve the concentration of zhongshengmycin, paclobutrazol, boscalid, and pyridaben, whose fluorescences are overlapping. The R squares of the models for zhongshengmycin, paclobutrazol, boscalid, and pyridaben were 0.9942, 0.9912, 0.9913, and 0.9847, respectively. Results indicated that fluorescence spectroscopy combined with multiple PLS models can be used to detect multiple kinds of pesticides in the water.

Determination and Modeling on Ultraviolet Light Degradation of Pyridaben Based on Fluorescence Spectrum.

AIMS AND OBJECTIVE: Pesticide residues seriously affect human health, so it is very important to study the degradation of pesticide residues for food safety. The degradation of pyridaben by ultraviolet (UV) irradiation was studied, the degradation characteristics and modeling were analyzed in this paper. This study was undertaken to fully reveal the degradation mechanism of UV irradiation for pyridaben residue and provided the evaluation method of degradation effect. MATERIALS AND METHODS: Firstly, the fluorescence spectra of pyridaben samples were measured by LS55 fluorescence photometer, and the relationship between pyridaben concentration and the fluorescence intensity of characteristic peak was established. Then, using UV irradiation approach, the pyridaben was degraded to different degrees by controlling the irradiation time. The degradation process was characterized according to the change of fluorescence characteristic peak intensity before and after degradation. The relationship between degradation time and fluorescence intensity was established at last. RESULTS: The results showed that the fluorescence characteristic peak of pyridaben was located at 356 nm. The pyridaben content prediction model function was obtained with the correlation coefficient of 0.9989 and the average recovery of 99.70%. The relative standard deviation (RSD%), the limit of detection (LOD) and the limit of quantity (LOQ) was 1.71%, 0.0058 ug/ml and 0.0193 ug/ml, respectively. The exponential function model between UV degradation time and fluorescence intensity was obtained, the corresponding correlation coefficient was 0.9991, and the average recovery was 100.49%. CONCLUSION: UV light irradiation can effectively degrade pyridaben, degradation process can be characterized by the change of fluorescence intensity, and the degradation model was tested to be accurate.

Dual-model analysis for improving the discrimination performance of human and nonhuman blood based on Raman spectroscopy.

The discrimination accuracy for human and nonhuman blood is important for customs inspection and forensic applications. Recently, Raman spectroscopy has shown effectiveness in analyzing blood droplets and stains with an excitation wavelength of 785 nm. However, the discrimination of liquid whole blood in a vacuum blood tube using Raman spectroscopy, which is a form of noncontact and nondestructive detection, has not been achieved. An excitation wavelength of 532 nm was chosen to avoid the fluorescent background of the blood tube, at the cost of reduced spectroscopic information and discrimination accuracy. To improve the accuracy and true positive rate (TPR) for human blood, a dual-model analysis method is proposed. First, model 1 was used to discriminate human-unlike nonhuman blood. Model 2 was then used to discriminate human-like nonhuman blood from the "human blood" obtained by model 1. A total of 332 Raman spectra from 10 species were used to build and validate the model. A blind test and external validation demonstrated the effectiveness of the model. Compared with the results obtained by the single partial least-squares model, the discrimination performance was improved. The total accuracy and TPR, which are highly important for practical applications, increased to 99.1% and 97.4% from 87.2% and 90.6%, respectively.

Error analysis of the spectral shift for partial least squares models in Raman spectroscopy.

Raman spectroscopy paired with the partial least squares (PLS) method is commonly used for quantitative or qualitative analysis of complex samples. However, spectral shift induced by different Raman spectroscopy, different environment or different measured time will decrease the accuracy of the PLS model. In this work, the processing algorithms that improve the accuracy by removing the noise, background and varying sources of other spectral interference were first reviewed. The error induced by the spectral shift was analyzed and the formulas of the error were derived. The formulas were then used to calculate the theoretical error in the example of discriminating human and nonhuman blood. A comparison of the actual errors obtained from the mathematical method and experiment with the theoretical value demonstrated the effectiveness of the equation. The compensation for nonhuman blood according to the average error demonstrated the improvement of the accuracy. Finally, the non-uniform sampling of the Raman shift by charge-coupled device (CCD) was considered in the error equation. An accurate error equation was obtained. This work could help improve the stability of PLS models in the case of the spectral shift of the spectrometer in Raman spectroscopy.

Discrimination of human and nonhuman blood using Raman spectroscopy with self-reference algorithm.

We report a self-reference algorithm to discriminate human and nonhuman blood by calculating the ratios of identification Raman peaks to reference Raman peaks and choosing appropriate threshold values. The influence of using different reference peaks and identification peaks was analyzed in detail. The Raman peak at 1003 cm-1 was proved to be a stable reference peak to avoid the influencing factors, such as the incident laser intensity and the amount of sample. The Raman peak at 1341 cm-1 was found to be an efficient identification peak, which indicates that the difference between human and nonhuman blood results from the C-H bend in tryptophan. The comparison between self-reference algorithm and partial least square method was made. It was found that the self-reference algorithm not only obtained the discrimination results with the same accuracy, but also provided information on the difference of chemical composition. In addition, the performance of self-reference algorithm whose true positive rate is 100% is significant for customs inspection to avoid genetic disclosure and forensic science.