Spectral Prediction Research Articles

Parameter Estimation in Spectral Resolution Enhancement Based on Forward-Backward Linear Prediction Total Least Square Method.

In a Fourier transform infrared (IR) spectrometer, the Michelson interference signal extrapolation method based on linear prediction is often used to improve spectral resolution. In this method, an autoregressive (AR) model is established for the Michelson interference signal in the spectrometer. Once the AR model parameters are determined, the AR process is predictable. The interference signal can be used to figure out the AR model's parameters. Based on this, the AR model can be used to extrapolate the interference signal to improve the spectral resolution. In this paper, the forward-backward linear prediction total least squares (FB-TLS) method is proposed to estimate the parameters of the AR model. The parameters that are estimated are used to improve the IR spectral resolution. By simulating different order and signal-to-noise ratio situations, the effects of the Burg, the least square, and the FB-TLS parameter estimation methods on spectral resolution enhancement are studied. The simulation results demonstrate that the FB-TLS parameter estimation method can effectively suppress noise and avoid spurious peaks. The experimental results demonstrate that the FB-TLS parameter estimation method is effective for spectral resolution enhancement technology based on linear prediction. When the FB-TLS method is used to enhance NH3 IR spectral resolution from 2 cm-1 to 1 cm-1, the spectral prediction error in the NH3 characteristic band is only 0.21% compared with the measured NH3 spectrum, whose spectral resolution is 1 cm-1.

Improvement of the prediction of a visual apple ripeness index under seasonal variation by NIR spectral model correction

Apple ripeness assessment is essential to ensure its post-harvest commercial value, and the visible/near-infrared(NIR) spectral models that are effective in achieving this goal are prone to failure due to seasonal or instrumental factors. This study has proposed a visual ripeness index (VRPI) determined by parameters such as soluble solids, titratable acids, etc., which vary during the ripening period of the apple. The R and RMSE of the index prediction model based on the 2019 sample were 0.871 to 0.913 and 0.184 to 0.213 respectively. The model failed to predict the next two years of the sample, which was effectively addressed by model fusion and correction. For the 2020 and 2021 samples, the revised model improves R by 6.8% and 10.6% and reduces RMSE by 52.2% and 32.2% respectively. The results showed that the global model is adapted to the correction of the VRPI spectral prediction model under seasonal variation.

Comparative Evaluation of Electron Ionization Mass Spectral Prediction Methods.

During the past decade promising methods for computational prediction of electron ionization mass spectra have been developed. The most prominent ones are based on quantum chemistry (QCEIMS) and machine learning (CFM-EI, NEIMS). Here we provide a threefold comparison of these methods with respect to spectral prediction and compound identification. We found that there is no unambiguous way to determine the best of these three methods. Among other factors, we find that the choice of spectral distance functions play an important role regarding the performance for compound identification.

Nonlinear model prediction of needle chlorophyll content of Picea koraiensis Nakai at different needle ages based on hyperspectral features

Pigment content is a critical assessment indicator in the study of plant physiological metabolism, stress resistance, ornamental characteristics, and forest health. Spectral imaging technology is widely used for rapid and non-destructive determination of plant physicochemical parameters. To address the shortcomings of previous models of spectral reflectance prediction of chlorophyll content of needles only from the perspective of traditional algorithms and ignoring physical models, this research integrates variable complexity and refined classification of physical models to validate the increased accuracy of both the conventional partial least squares (PLS) method and the traditional neural network algorithm. The results of the conifer chlorophyll models of Picea koraiensis Nakai with different needle ages based on spectral reflectance and vegetation index parameters showed that the improved nonlinear state transition algorithm-backpropagation (STA-BP) neural network model approach (R2 of 0.73–0.89) and the nonlinear Stacking partial least squares (Stacking-PLS) model approach (R2 of 0. 67–0.85) is slightly more robust than the traditional algorithms nonlinear BP model (R2 of 0.63–0.82) and linear PLS model (R2 of 0.60–0.76). This finding suggests that the nonlinear fitting of chlorophyll content in needles of different needle ages in P. koraiensis Nakai surpasses the traditional linear model fitting methodology. Furthermore, the model fitting of chlorophyll content in conifers of different needle ages outperforms the mixed P. koraiensis Nakai model, suggesting that chlorophyll models using needle refinement classification help to improve model robustness. This study provides data and theoretical support for rapid and non-invasive characterization of physiological and biochemical properties of needles of different needle ages using spectral imaging techniques to predict growth and community structure productivity of forest trees in the coming years.

Quality Assessment and Ripeness Prediction of Table Grapes Using Visible-Near-Infrared Spectroscopy.

Ripeness significantly affects the commercial values and sales of fruits. In order to monitor the change of grapes' quality parameters during ripening, a rapid and nondestructive method of visible-near-infrared spectral (Vis-NIR) technology was utilized in this study. Firstly, the physicochemical properties of grapes at four different ripening stages were explored. Data evidenced increasing color in redness/greenness (a*) and Chroma (C*) and soluble solids (SSC) content and decreasing values in color of lightness (L*), yellowness/blueness (b*) and Hue angle (h*), hardness, and total acid (TA) content as ripening advanced. Based on these results, spectral prediction models for SSC and TA in grapes were established. Effective wavelengths were selected by the competitive adaptive weighting algorithm (CARS), and six common preprocessing methods were applied to pretreat the spectra data. Partial least squares regression (PLSR) was applied to establish models on the basis of effective wavelengths and full spectra. The predictive PLSR models built with full spectra data and 1st derivative preprocessing provided the best values of performance parameters for both SSC and TA. For SSC, the model showed the coefficients of determination for calibration (RCal2) and prediction (RPre2) set of 0.97 and 0.93, respectively, the root mean square error for calibration set (RMSEC) and prediction set (RMSEP) of 0.62 and 1.27, respectively; and the RPD equal to 4.09. As for TA, the optimum values of RCal2, RPre2, RMSEC, RMSEP and RPD were 0.97, 0.94, 0.88, 1.96 and 4.55, respectively. The results indicated that Vis-NIR spectroscopy is an effective tool for the rapid and non-destructive detection of SSC and TA in grapes.

Mistle: bringing spectral library predictions to metaproteomics with an efficient search index.

Deep learning has moved to the forefront of tandem mass spectrometry-driven proteomics and authentic prediction for peptide fragmentation is more feasible than ever. Still, at this point spectral prediction is mainly used to validate database search results or for confined search spaces. Fully predicted spectral libraries have not yet been efficiently adapted to large search space problems that often occur in metaproteomics or proteogenomics. In this study, we showcase a workflow that uses Prosit for spectral library predictions on two common metaproteomes and implement an indexing and search algorithm, Mistle, to efficiently identify experimental mass spectra within the library. Hence, the workflow emulates a classic protein sequence database search with protein digestion but builds a searchable index from spectral predictions as an in-between step. We compare Mistle to popular search engines, both on a spectral and database search level, and provide evidence that this approach is more accurate than a database search using MSFragger. Mistle outperforms other spectral library search engines in terms of run time and proves to be extremely memory efficient with a 4- to 22-fold decrease in RAM usage. This makes Mistle universally applicable to large search spaces, e.g. covering comprehensive sequence databases of diverse microbiomes. Mistle is freely available on GitHub at https://github.com/BAMeScience/Mistle.

Circular dichroism simulations of chiral buckybowls by means curvature analyses

A detailed understanding and interpretation of chiral properties of molecular systems, especially in condensed phase, often requires computational models that allow their structural and electronic features to be connected to the observed experimental spectra. The present paper is focused on modelling the circular dichroism spectra of chiral buckybowls, combining topological aspects and the density functional theory. For the first time Ball Pivoting Algorithm was proposed to hook up the chemical topology to the DFT through the surface reconstruction. Particularly, the gaussian curvature of a constructed probe set of corannulene and sumanene derivatives was used as discriminant parameter to benchmark a list of 10 functionals (B3LYP, B97D, M06-2X, HSEH1PBE, wB97XD, CAM-B3LYP, LC-wPBE, TPSSTPSS, mPW1PW91 and APFD). The latter provide to be noticeably accurate to reproduce the curvature effect of the considered molecules. A TD-DFT/BOMD mixed approach provided a comprehensive overview of the spectral chiral pattern prediction trends when multiple DFT functionals are scanned. The preliminary topological analysis efforts were then recompensed with the very precise computed CD spectra, again APFD confirmed as the leader functional, this time for TD-DFT vertical transition calculations. Therefore, we strongly recommend the use of the of dispersion embedded APFD functional coupled with the 6–311++G(2d,2p) basis set for the computation of the functionalized chiral buckybowls ECD spectra.© 2017 Elsevier Inc. All rights reserved.

Terahertz Spectroscopy Characterization and Prediction of the Aging Degree of Polyethylene Pipes Based on PLS.

Polyethylene (PE) is widely used in pipeline transportation owing to its excellent corrosion resistance, good stability, and ease of processing. As organic polymer materials, PE pipes inevitably undergo different degrees of aging during long-term use. In this study, terahertz time-domain spectroscopy was used to study the spectral characteristics of PE pipes with different degrees of photothermal aging, and the variation in the absorption coefficient with aging time was obtained. The absorption coefficient spectrum was extracted using uninformative variable elimination (UVE), successive projections algorithm (SPA), competitive adaptive reweighted sampling (CARS), and random frog RF spectral screening algorithms, and the spectral slope characteristics of the aging-sensitive band were selected as the evaluation indices of the degree of PE aging. Based on this, a partial least squares aging characterization model was established to predict white PE80, white PE100 and black PE100 pipes with different aging degrees. The results showed that the prediction accuracy of the absorption coefficient spectral slope feature prediction model for the aging degree of different types of pipes was greater than 93.16% and the verification set error was within 13.5 h.

The Transferability of Spectral Grain Yield Prediction in Wheat Breeding across Years and Trial Locations.

Grain yield (GY) prediction based on non-destructive UAV-based spectral sensing could make screening of large field trials more efficient and objective. However, the transfer of models remains challenging, and is affected by location, year-dependent weather conditions and measurement dates. Therefore, this study evaluates GY modelling across years and locations, considering the effect of measurement dates within years. Based on a previous study, we used a normalized difference red edge (NDRE1) index with PLS (partial least squares) regression, trained and tested with the data of individual dates and date combinations, respectively. While strong differences in model performance were observed between test datasets, i.e., different trials, as well as between measurement dates, the effect of the train datasets was comparably small. Generally, within-trials models achieved better predictions (max. R2 = 0.27-0.81), but R2-values for the best across-trials models were lower only by 0.03-0.13. Within train and test datasets, measurement dates had a strong influence on model performance. While measurements during flowering and early milk ripeness were confirmed for within- and across-trials models, later dates were less useful for across-trials models. For most test sets, multi-date models revealed to improve predictions compared to individual-date models.

Spectral Level Prediction Model of Ocean Ambient Noise Based on GA-LM-BP Neural Network

Spectral Level Prediction Model of Ocean Ambient Noise Based on GA-LM-BP Neural Network

TSWIFT: Tower Spectrometer on Wheels for Investigating Frequent Timeseries for high-throughput phenotyping of vegetation physiology

BackgroundRemote sensing instruments enable high-throughput phenotyping of plant traits and stress resilience across scale. Spatial (handheld devices, towers, drones, airborne, and satellites) and temporal (continuous or intermittent) tradeoffs can enable or constrain plant science applications. Here, we describe the technical details of TSWIFT (Tower Spectrometer on Wheels for Investigating Frequent Timeseries), a mobile tower-based hyperspectral remote sensing system for continuous monitoring of spectral reflectance across visible-near infrared regions with the capacity to resolve solar-induced fluorescence (SIF).ResultsWe demonstrate potential applications for monitoring short-term (diurnal) and long-term (seasonal) variation of vegetation for high-throughput phenotyping applications. We deployed TSWIFT in a field experiment of 300 common bean genotypes in two treatments: control (irrigated) and drought (terminal drought). We evaluated the normalized difference vegetation index (NDVI), photochemical reflectance index (PRI), and SIF, as well as the coefficient of variation (CV) across the visible-near infrared spectral range (400 to 900 nm). NDVI tracked structural variation early in the growing season, following initial plant growth and development. PRI and SIF were more dynamic, exhibiting variation diurnally and seasonally, enabling quantification of genotypic variation in physiological response to drought conditions. Beyond vegetation indices, CV of hyperspectral reflectance showed the most variability across genotypes, treatment, and time in the visible and red-edge spectral regions.ConclusionsTSWIFT enables continuous and automated monitoring of hyperspectral reflectance for assessing variation in plant structure and function at high spatial and temporal resolutions for high-throughput phenotyping. Mobile, tower-based systems like this can provide short- and long-term datasets to assess genotypic and/or management responses to the environment, and ultimately enable the spectral prediction of resource-use efficiency, stress resilience, productivity and yield.

Spectral Quantitative Analysis and Research of Fusarium Head Blight Infection Degree in Wheat Canopy Visible Areas

Obtaining complete and consistent spectral images of wheat ears in the visible areas of in situ wheat canopies poses a significant challenge due to the varying growth posture of wheat. Nevertheless, detecting the presence and degree of wheat Fusarium head blight (FHB) in situ is critical for formulating measures that ensure stable grain production and supply while promoting green development in agriculture. In this study, a spectral quantitative analysis model was developed to evaluate the infection degree of FHB in an in situ wheat canopy’s visible areas. To achieve this, a spectral acquisition method was used to evaluate the infection degree of FHB in a wheat canopy’s visible areas. Hyperspectral images were utilized to obtain spectral data from healthy and mildly, moderately, and severely infected wheat ear canopies. The spectral data were preprocessed, and characteristic wavelengths were extracted using twelve types of spectral preprocessing methods and four types of characteristic wavelength extraction methods. Subsequently, sixty-five spectral quantitative prediction models for the infection degree of FHB in the in situ wheat canopy visible areas were established using the PLSR method, based on the original spectral data, preprocessed spectral data, original spectral characteristic wavelengths extracted data, and preprocessed spectral characteristic wavelengths extracted data. Comparative analysis of the models indicated that the MMS + CARS + PLSR model exhibited the best prediction effect and could serve as the spectral quantitative analysis model for the evaluation of the infection degree of FHB in an in situ wheat canopy’s visible areas. The model extracted thirty-five characteristic wavelengths, with a modeling set coefficient of determination (R2) of 0.9490 and a root-mean-square error (RMSE) of 0.2384. The testing set of the coefficient of determination (R2) was 0.9312, with a root-mean-square error (RMSE) of 0.2588. The model can facilitate the spectral quantitative analysis of the infection degree of FHB in the in situ wheat canopy visible areas, thereby aiding in the implementation of China’s targeted poverty alleviation and agricultural power strategy.

Carbon Stock Prediction in Managed Forest Ecosystems Using Bayesian and Frequentist Geostatistical Techniques and New Generation Remote Sensing Metrics

The study compares the performance of a hierarchical Bayesian geostatistical methodology with a frequentist geostatistical approach, specifically, Kriging with External Drift (KED), for predicting C stock using prediction aides from the Landsat-8 and Sentinel-2 multispectral remote sensing platforms. The frequentist geostatistical approach’s reliance on the long-run frequency of repeated experiments for constructing confidence intervals is not always practical or feasible, as practitioners typically have access to a single dataset due to cost constraints on surveys and sampling. We evaluated two approaches for C stock prediction using two new generation multispectral remote sensing datasets because of the inherent uncertainty characterizing spatial prediction problems in the unsampled locations, as well as differences in how the Bayesian and frequentist geostatistical paradigms handle uncertainty. Information on C stock spectral prediction in the form of NDVI, SAVI, and EVI derived from multispectral remote sensing platforms, Landsat-8 and Sentinel-2, was used to build Bayesian and frequentist-based C stock predictive models in the sampled plantation forest ecosystem. Sentinel-2-based C stock predictive models outperform their Landsat-8 counterparts using both the Bayesian and frequentist inference approaches. However, the Bayesian-based Sentinel-2 C stock predictive model (RMSE = 0.17 MgCha−1) is more accurate than its frequentist-based Sentinel-2 (RMSE = 1.19 MgCha−1) C stock equivalent. The Sentinel-2 frequentist-based C stock predictive model gave the C stock prediction range of 1 ≤ MgCha−1 ≤ 290, whilst the Sentinel-2 Bayesian-based C stock predictive model resulted in the prediction range of 1 ≤ MgCha−1 ≤ 285. However, both the Bayesian and frequentist C stock predictive models built with the Landsat-8 sensor overpredicted the sampled C stock because the range of predicted values fell outside the range of the observed C stock values. As a result, we recommend and conclude that the Bayesian-based C stock prediction method, when it is combined with high-quality remote sensing data such as that of Sentinel-2, is an effective inferential statistical methodology for reporting C stock in managed plantation forest ecosystems.

Improved Support Vector-Recurrent Neural Network with Optimal Feature Selection-based Spoken Language Identification System

Objective: Spoken language identification being the fore-front of language recognition tasks and most significant medium of communication has to be enhanced in order to improve the accuracy of recently developed spoken language recognition systems. The purpose of this paper is to enhance the Spoken Language Identification (SLID) model using hybrid machine learning with deep learning model for regionally spoken languages of Jammu & Kashmir (JK) and Ladakh. Method: Initially, the speech signals of different languages of JK and Ladakh are manually collected from diverse sources, and it is preprocessed using Spectral Noise Gate (SNG) filtering technique. Once the speech signals are pre-processed, the feature extraction is performed by the cepstral features like Mel-frequency Cepstral Coefficients (MFCCs), Relative Spectral Transform-Perceptual Linear Prediction (RASTA-PLP), and spectral features like spectral roll off, spectral flatness. Findings: From this feature extraction, the length of the feature vector seems to be long, and it is required to reduce the feature length. Hence, optimal feature selection is accomplished using the new meta-heuristic algorithm termed Adaptive Distance-based Tunicate Swarm Algorithm (AD-TSA) by considering the minimum correlation as objective. Finally, the language identification is handled by the hybrid classifier termed Improved Support Vector Machine-Recurrent Neural Network (ISVM-RNN). Novelty: The identification learning algorithm is enhanced by the AD-TSA by considering the minimum correlation as objective among features in order to get minimum number of features that are sufficient for language identification process. The efficiency of the proposed hybrid approach is validated by simulating the experiment on a user-defined language database of JK and Ladakh speech signals in the working platform of Python. Keywords: Language Identification; Kashmir Languages; Optimal Feature Selection; Improved Support Vector MachineRecurrent Neural Network; Adaptive DistanceBased Tunicate Swarm Algorithm

Physics-informed dynamic mode decomposition

In this work, we demonstrate how physical principles—such as symmetries, invariances and conservation laws—can be integrated into thedynamic mode decomposition(DMD). DMD is a widely used data analysis technique that extracts low-rank modal structures and dynamics from high-dimensional measurements. However, DMD can produce models that are sensitive to noise, fail to generalize outside the training data and violate basic physical laws. Our physics-informed DMD (piDMD) optimization, which may be formulated as a Procrustes problem, restricts the family of admissible models to a matrix manifold that respects the physical structure of the system. We focus on five fundamental physical principles—conservation, self-adjointness, localization, causality and shift-equivariance—and derive several closed-form solutions and efficient algorithms for the corresponding piDMD optimizations. With fewer degrees of freedom, piDMD models are less prone to overfitting, require less training data, and are often less computationally expensive to build than standard DMD models. We demonstrate piDMD on a range of problems, including energy-preserving fluid flow, the Schrödinger equation, solute advection-diffusion and three-dimensional transitional channel flow. In each case, piDMD outperforms standard DMD algorithms in metrics such as spectral identification, state prediction and estimation of optimal forcings and responses.

Monitoring Cadmium Content in the Leaves of Field Pepper and Eggplant in a Karst Area Using Hyperspectral Remote Sensing Data

The ability to quickly and non-destructively monitor the cadmium (Cd) content in agricultural crops is the basic premise of effective prevention and control of Cd contamination in agricultural products. Hyperspectral technology provides a solution for this issue. The potential capability for the spectral prediction of the Cd content in the leaves of pepper and eggplant in the field was explored, and a spectral prediction model of the Cd content in these leaves was established. In this study, based on the indoor spectrum, the sensitive wavebands for predicting the Cd content in leaves were determined preliminarily by correlation analysis. Partial least squares regression (PLSR) and support vector machine regression (SVMR) were used to establish spectral prediction models, and the final sensitive wavebands were determined by the size of the model index. The results show that the SVMR model exhibited higher prediction accuracy than the PLSR model. The RPDp (relative percent different of prediction set) values of the best SVMR prediction models for the pepper leaves and the eggplant leaves were 1.82 and 1.49, respectively. The values of Rp2 (coefficient of determination of prediction set), which can quantitatively estimate the Cd content in leaves, were 0.897 (p < 0.01) and 0.726 (p < 0.01), respectively. This study demonstrated that the leaf spectra of pepper and eggplant in the field can be used to predict the Cd content in leaves, providing a reference for monitoring the Cd content in the fruits of pepper and eggplant in the future.

A comprehensive deep learning method for empirical spectral prediction and its quantitative validation of nano-structured dimers

Nanophotonics exploits the best of photonics and nanotechnology which has transformed optics in recent years by allowing subwavelength structures to enhance light-matter interactions. Despite these breakthroughs, design, fabrication, and characterization of such exotic devices have remained through iterative processes which are often computationally costly, memory-intensive, and time-consuming. In contrast, deep learning approaches have recently shown excellent performance as practical computational tools, providing an alternate avenue for speeding up such nanophotonics simulations. This study presents a DNN framework for transmission, reflection, and absorption spectra predictions by grasping the hidden correlation between the independent nanostructure properties and their corresponding optical responses. The proposed DNN framework is shown to require a sufficient amount of training data to achieve an accurate approximation of the optical performance derived from computational models. The fully trained framework can outperform a traditional EM solution using on the COMSOL Multiphysics approach in terms of computational cost by three orders of magnitude. Furthermore, employing deep learning methodologies, the proposed DNN framework makes an effort to optimise design elements that influence the geometrical dimensions of the nanostructure, offering insight into the universal transmission, reflection, and absorption spectra predictions at the nanoscale. This paradigm improves the viability of complicated nanostructure design and analysis, and it has a lot of potential applications involving exotic light-matter interactions between nanostructures and electromagnetic fields. In terms of computational times, the designed algorithm is more than 700 times faster as compared to conventional FEM method (when manual meshing is used). Hence, this approach paves the way for fast yet universal methods for the characterization and analysis of the optical response of nanophotonic systems.

Convolutional neural networks used for random structure SPP gratings spectral response prediction.

Data-driven design approaches based on deep learning have been introduced into nanophotonics to reduce time-consuming iterative simulations, which have been a major challenge. Here, we report a convolutional neural network (CNN) used to perform the prediction of surface plasmon polariton (SPP) grating output spectra, which is not limited by predefined shapes. For a random given structure, the network can output spectra with effective prediction, so that the simulation results are in excellent agreement with the network prediction results. Compared with the traditional finite-difference time-domain (FDTD) method, the CNN model proposed in this Letter has absolute advantages in speed. Previous studies often used a regular device structure to modify its parameters for prediction; the random structure design method adopted in this Letter also provides a new, to the best of knowledge, idea for device design.

COVID-19 Detection via Fusion of Modulation Spectrum and Linear Prediction Speech Features

The coronavirus disease 2019 (COVID-19) pandemic has drastically impacted life around the globe. As life returns to pre-pandemic routines, COVID-19 testing has become a key component, assuring that travellers and citizens are free from the disease. Conventional tests can be expensive, time-consuming (results can take up to 48h), and require laboratory testing. Rapid antigen testing, in turn, can generate results within 15-30 minutes and can be done at home, but research shows they achieve very poor sensitivity rates. In this paper, we propose an alternative test based on speech signals recorded at home with a portable device. It has been well-documented that the virus affects many of the speech production systems (e.g., lungs, larynx, and articulators). As such, we propose the use of new modulation spectral features and linear prediction analysis to characterize these changes and design a two-stage COVID-19 prediction system by fusing the proposed features. Experiments with three COVID-19 speech datasets (CSS, DiCOVA2, and Cambridge subset) show that the two-stage feature fusion system outperforms the benchmark systems of CSS and Cambridge datasets while maintaining lower complexity compared to DL-based systems. Furthermore, the two-stage system demonstrates higher generalizability to unseen conditions in a cross-dataset testing evaluation scheme. The generalizability and interpretability of our proposed system demonstrate the potential for accessible, low-cost, at-home COVID-19 testing.

Non-destructive Detection of Fatty Acid Content of Camellia Seed Based on Hyperspectral.

As a unique traditional vegetable oil in China, camellia seed oil has very high edible value. Camellia seed kernel is mainly composed of fatty acids, which not only determines the oil yield of camellia seed, but also exert an important impact on the storage performance of camellia seed. In order to quickly and accurately determine the fatty acid content of camellia seed, this paper took camellia seed as the research object, used hyperspectral technology to determine the fatty acid content of camellia seed, and establishes a spectral model. 8 pretreatment methods, such as Savitzky-Golay smoothing, normalization, baseline correction, multivariate scattering correction, standard normal variable transformation, detrending algorithm, first derivative and second derivative, were adopted in this paper. The spectral prediction model of fatty acid content in camellia seed was established by combining 4 modeling methods: principal components regression (PCR), partial least square regression (PLSR), back propagation neural network (BP), radial basis function neural network (RBF). The optimal prediction model was selected by comparing the coefficient of determination (R2) and root mean square error (RMSE) of various models. The results showed that the spectral sensitive bands with high correlation coefficients (r) were 410-420 nm, 450-460 nm, 490-510 nm, 545-580 nm, 845-870 nm and 905-925 nm, respectively. The r obtained by MSC pretreatment of spectral data was the largest. The data obtained by 8 different pretreatment methods combined with RBF neural network model was the best, in which the average value of coefficient of determination (RC2) in the calibration set was 0.8654, and the root mean square error of calibration (RMSEC) was 0.0777; the average value of coefficient of determination (R2P) and root mean square error of prediction (RMSEP) in the prediction set model were 0.8437 and 0.0827, respectively. It could be seen that the best accuracy could be achieved by MSC pretreatment combined with RBF neural network modeling. This paper can provide reference for rapid nondestructive detection of fatty acid content in camellia seed by hyperspectral technology.