Hyperspectral Reflectance Imaging Research Articles

Monitoring soil nutrients using machine learning based on UAV hyperspectral remote sensing

ABSTRACT Unmanned aerial vehicles (UAV) are rapidly evolving experimental platforms that play an important role in remote sensing. In this study, we investigated a machine learning method for monitoring soil nutrient content using UAV hyperspectral remote sensing. We employed machine learning techniques for feature extraction and soil hyperspectral information modelling. In contrast to traditional mathematical transformation methods, we adopted a combination of random forest and differential evolution algorithms to rank the weights of individual hyperspectral data, thereby obtaining a series of spectral feature subsets for soil organic matter, total nitrogen and available phosphorus and potassium. Furthermore, the analytic hierarchy process was used for weight analysis, and the characteristic bands of the four soil nutrients were successfully extracted. Next, a quantitative inversion model based on a back-propagation (BP) neural network was established to estimate soil nutrient content, with determination coefficients higher than 0.7 and 0.6 for the modelling and verification sets, respectively. The relative percent difference values were greater than 2, among which the highest was for available potassium, with determination coefficients of 0.95 and 0.84 for the modelling and verification sets, respectively. In addition, visualization distribution maps of soil nutrients were obtained by combining the BP model and original reflectance hyperspectral images, and the comparisons of content histograms showed a relatively consistent distribution between the sampling and inversion points. The results verified the effectiveness of the combined machine learning method for large-scale and high-precision monitoring and visualization of soil nutrient contents.

Monitoring soil nutrients using machine learning based on UAV hyperspectral remote sensing

ABSTRACT Unmanned aerial vehicles (UAV) are rapidly evolving experimental platforms that play an important role in remote sensing. In this study, we investigated a machine learning method for monitoring soil nutrient content using UAV hyperspectral remote sensing. We employed machine learning techniques for feature extraction and soil hyperspectral information modelling. In contrast to traditional mathematical transformation methods, we adopted a combination of random forest and differential evolution algorithms to rank the weights of individual hyperspectral data, thereby obtaining a series of spectral feature subsets for soil organic matter, total nitrogen and available phosphorus and potassium. Furthermore, the analytic hierarchy process was used for weight analysis, and the characteristic bands of the four soil nutrients were successfully extracted. Next, a quantitative inversion model based on a back-propagation (BP) neural network was established to estimate soil nutrient content, with determination coefficients higher than 0.7 and 0.6 for the modelling and verification sets, respectively. The relative percent difference values were greater than 2, among which the highest was for available potassium, with determination coefficients of 0.95 and 0.84 for the modelling and verification sets, respectively. In addition, visualization distribution maps of soil nutrients were obtained by combining the BP model and original reflectance hyperspectral images, and the comparisons of content histograms showed a relatively consistent distribution between the sampling and inversion points. The results verified the effectiveness of the combined machine learning method for large-scale and high-precision monitoring and visualization of soil nutrient contents.

Fusing spectral and spatial features of hyperspectral reflectance imagery for differentiating between normal and defective blueberries

Effective defect detection of blueberries is important to ensuring supplies of high-quality products to the fresh market. In this study, hyperspectral reflectance imaging with machine learning was evaluated for discriminating between defective and normal blueberries. Fresh blueberries hand-harvested were scanned in the wavelength range of 400–1000 nm. An image analysis pipeline was developed to segment individual blueberries and extract mean spectra and spatial features. Defective blueberries were found to have lower near-infrared reflectance than sound samples, and spectral features produced a better separation between defective and sound samples than the spatial features in the scattering plots of their first two principal components. Nine types of machine learning models were built for classifying defective and sound samples using the spectral and spatial features separately as well as their concatenation. The regularized linear discriminant analysis (RLDA) model trained on the spectral features achieved the best overall accuracy of 95.7 %, as opposed to the best accuracy of 85.3 % based on spatial features, which was obtained by LDA. Simply concatenating spectral and spatial features did not improve over modeling using spectral or spatial features alone. A model ensemble strategy integrating the spectral features-based RLDA and the spatial features-based LDA resulted in a statistically significant improvement in the overall accuracy to 96.6 %. Model-level feature integration offers an effective means for improving the discrimination between defective and normal blueberries. Both the hyperspectral data11https://doi.org/10.5281/zenodo.11200576. and the software programs22https://github.com/vicdxxx/Blueberry-Defect-Detection-by-Hyperspectral-Imaging. of this study are made publicly available.

Research on Hyperspectral Surface Reflectance Dataset of Typical Ore Concentration Area in Hami Remote Sensing Test Field

Abstract. Surface reflectance data is the basic data source for the hyperspectral parametric remote sensing products and remote sensing quantitative application, which is widely used in various application fields such as natural resources and ecological environment monitoring. At present, multispectral data takes the leading role among the common land surface reflectance datasets and the reflectance data mainly involves the types of ground objects such as farmland, forest land, water body, soil, etc., while the datasets relatively less targets the types of rock and mineral surface objects, yet especially the reflectance datasets with the combination of time series and multi-scale satellite-earth are even more scarce. In order to better promote the application of hyperspectral surface reflectance and explore the advantages of joint application of satellite-earth multi-scale reflectance data, on the basis of field-measured rock and mineral target spectral, a comprehensive surface reflectance dataset was generated by using domestically produced hyperspectral satellite data as the data source in this study, mainly focusing on the typical ore concentration area in the Hami Remote Sensing test field in Xinjiang. The dataset includes multi-period hyperspectral satellite surface reflectance images, field measured rock and mineral spectral data, and multi-period sub-pixel spectral data collected based on ground spectral measured points, which can provide significant support for the research and development and accuracy verification as well as performance evaluation of algorithms such as surface reflectance inversion, mineral identification and ground object classification.

Quantification of early bruises on blueberries using hyperspectral reflectance imaging coupled with band ratio and improved multi-threshold coati optimization algorithm method

Bruises resulting from mechanical collisions during picking, packing, and transportation lead to a decline in the quality of blueberry fruits. Therefore, performing a quantitative evaluation in the early stages of fruit bruising is essential for achieving the quality classification of blueberries. The primary aim of this study was to precisely quantify internal damage in blueberries by employing a combination of band ratios and an improved multi-threshold Coati Optimization Algorithm (MT-COA) method. Initially, hyperspectral reflectance images (ranging from 950 to 1650 nm) of both sound and bruised blueberries were acquired 30 min after damage. Secondly, to reduce data dimensionality, data processing utilized principal component analysis and band ratio techniques, and a comparative analysis of these two methods was performed. Subsequently, the band ratio images R1112.6/1470.9 were segmented into three classes, encompassing bruised tissue, sound tissue, and background, using global threshold, Otsu method, watershed segmentation algorithm (WSA), MT-COA method, and improved MT-COA method. Finally, the intersection over Union (IoU) metric was employed to assess the segmentation effectiveness. The bruise ratio index (BRI) was calculated to assess the damage rate of the fruit. The results indicate that by utilizing hyperspectral reflectance imaging in conjunction with band ratio and improved MT-COA method, sound and bruised areas can be segmented from the fruit as early as 30 min after damage. The average IoU accuracy is 70.1%. The model shows the potential to quantify early bruises of blueberries during the postharvest period.

Medición del marmoleado de carne de res usando imágenes espectrales: Un enfoque de regresión lineal múltiple

This study aimed at measuring beef marbling scores in an objective and simple manner through spectral imaging and multiple linear regression (MLR). Beef marbling prediction by hyperspectral imaging and partial least squares regression (PLSR) was analyzed to calibrate and evaluate an MLR model with a few selected wavelengths. Data came from 44 beef samples and consisted of their spectral signatures (75 wavelengths) from hyperspectral reflectance images (400-1000 nm) and their marbling scores assigned by evaluators. The wavelengths that presented regression coefficients with the highest absolute values in the PLSR model, were used to calibrate the MLR model by a backward stepwise approach (p < 0.05). The coefficient of determination for prediction (R2p) and the standard error of prediction (SEP) were evaluated. The MLR model was suitable for practical use because it required only 12 wavelengths for reliable predictions (R2p = 0.824 > 0.8; SEP = 11.4% < 15%). A model is proposed for the objective and simple measurement of beef marbling score using multispectral imaging technology.

IRIS: A novel integrated instrument for co‐registered MA‐XRF mapping and VNIR‐SWIR hyperspectral imaging

AbstractThe combination of complementary techniques for materials analysis can play a key role in both art conservation and academic research. Nowadays, the correlation of x‐ray fluorescence (XRF) with hyperspectral reflectance imaging in the visible and infrared region has become a valuable tool for palette identification, painting techniques studies and for the diagnostic support dedicated to restoration and conservation. Moreover, both techniques enable researchers to reveal fascinating underpaintings, “pentimenti”, or even preparatory drawings offering new details on the creative process of the artist. This background has been a strong motivation for the development of a new multimodal tool for art and conservation: IRIS. IRIS is a mobile and reconfigurable scanner designed to address a wide range of demanding application, exploiting the opportunities given by simultaneous MA‐XRF and hyperspectral reflectance scanning in the visible‐near‐infrared (VNIR) and short‐wave‐infrared (SWIR) range from 400 to 2500 nm. The system has been designed for in‐situ, fast and non‐invasive scanning of the sample without compromising spectral resolution and high throughput performance. The scanner acquires co‐registered XRF/VNIR‐SWIR data, thus allowing the user to obtain the maximum profit from their possible correlated information: the two techniques can provide enhanced or complementary information on the same spot of analysis with minimum effort in terms of data processing and no need for spatial alignment. In the present work, the qualitative and quantitative performance of IRIS are explored, together with the presentation of in‐lab analysis on reference samples and a brief insight on a real case‐study.

Classification of the Nutritional Status of Peach Trees Using Indexes from Hyperspectral Images

This paper presents a procedure for the comparison of two technologies developed to classify peach trees according to their nutritional status. The first technology uses the leaf SPAD-502 meter value to characterize tree classes as indicated by agronomist experts: sound, intermediate, and strong chlorosis trees. It is used as a reference for the second technology, which uses a combination of two multispectral indexes computed from reflectance hyperspectral images. Specifically, R_NDVI = (R800 − R670)/(R800 + R670) and HyperSPAD = (R940/R650) are computed for each leaf pixel. An automated methodology is proposed that sets two optical thresholds (three hyperspectral categories) in view of the outliers according to a normal distribution, together with an iterative optimization of the bounding that determines the best assignment of trees to one of the three SPAD_502 levels of nutritional status, as required for practical agronomical purposes such as fertilization. The Chi 2 distribution is used to confirm the similarity of both nutritional classifications. These results encourage the use of on-board multispectral cameras to monitor the nutritional status of trees and to establish a more efficient fertilization strategy where inputs are applied according to individual status, with the consequent reduction in losses of fertilizers such as nitrogen to the atmosphere, soil, and water resulting from over-application.

Practical recommendations and limitations for pushbroom hyperspectral imaging of tree stems

In this short communication, we present a pilot study testing a new close-range sensing technology – a portable, pushbroom hyperspectral camera – in varying field conditions in forests. We evaluate how measurement conditions affect the in situ collection of stem bark spectra. In situ spectral libraries of woody elements are needed in, e.g., physically-based remote sensing applications, biodiversity mapping, and 3D vegetation modeling. Recent technological advancements bring portable and close-range capable sensors, such as small pushbroom imaging spectrometers, for consumer and research use. However, it is important to investigate the strengths and limitations of sensors utilizing pushbroom technology. Spectral measurements under forest canopies are challenging due to varying illumination conditions, which can have a significant effect on the quality of the data. We acquired hyperspectral reflectance images of Norway spruce (Picea abies (L.) Karst), Scots pine (Pinus sylvestris L.), and silver birch (Betula pendula Roth) stem bark directly in the forest. For each tree we collected reflectance images at 30-minute intervals throughout a day from a fixed view angle. The most significant change in the measured spectra occurred due to spatially varying irradiance between the white reference panel and the bark surface. The spatial variation of irradiance had the largest effect on data quality in visible and red-edge regions, and the smallest in near-infrared. In non-diffuse conditions, changes in irradiance were often unpredictable as clouds or canopy elements moved in and out of the direct solar beams. Diffuse overcast days with clouds can extend the time window for measurements, making it a practical choice for acquiring hyperspectral images of stem bark. We concluded that with a well-planned measurement set-up it is possible to improve the precision of in situ collected spectra of stem bark.

Carotenoid Content Estimation in Tea Leaves Using Noisy Reflectance Data

Quantifying carotenoid content in agriculture is essential for assessing crop nutritional value, improving crop quality, promoting human health, understanding plant stress responses, and facilitating breeding and genetic improvement efforts. Hyperspectral reflectance imaging is a nondestructive and rapid tool for estimating the carotenoid content. In spectrometer reflectance measurements, there are various sources of noise that can compromise the accuracy of carotenoid content estimations. Recently, various machine learning algorithms have been identified as robust against various types of noise, eliminating the need for denoising processes. Specifically, Cubist and the one-dimensional convolutional neural network (1D-CNN) have been used in evaluating vegetation properties based on reflectance data. We used regression models based on Cubist and 1D-CNN to estimate carotenoid content from reflectance data (the spectral resolution was resampled in 5 nm bands across the entire wavelength domain from 400 to 850 nm) with various degrees of Gaussian and spike noise added. The Cubist-based model was the most robust for this purpose: it achieved a ratio of performance to deviation of 1.41, a root mean square error of 1.11 µg/cm2, and a coefficient of determination (R2) of 0.496 when applied to reflectance data with a combination of Gaussian (mean: 0; variance: 0.04) and spike noise (density: 0.05; amplitude: 0.05).

An investigation of Bellini’s Transfiguration in the Capodimonte museum by means of XRF and visible reflectance hyperspectral imaging: Bellini’s handling of materials in the head and figure of the transfigured Christ

The Transfiguration by Giovanni Bellini in the Museo di Capodimonte (Naples) is an iconic image of a transcendental event at the heart of the Christian religion. Using non-destructive chemical imaging, and focusing on the central image of the transfigured Christ we have studied by what means and with what materials the artist represented this divine event. The results showed what is to date the earliest use of the pigment stibnite. The techniques employed demonstrate both continuity with Bellini’s earliest works (such as the combination of the use of shell-gold and organic, tinted layers), but also his knowledge of Netherlandish painting. This research has furthered our understanding of how he achieved the luminosity so characteristic of his translation of the divine into the materiality of paint.

The man who became a Blue Glass. Reflectance hyperspectral imaging discloses a hidden Picasso painting of the Blue period

Reflectance hyperspectral imaging (HSI) in the Visible and Near-IR (VNIR) to the Short Wave IR (SWIR) ranges is acknowledged as one of the most effective techniques for non-invasive examination of paintings. Recently, in some important museums worldwide a new trend has emerged towards a more systematic use of reflectance HSI, which is applied as advanced imaging technique to attain at once analysis and documentation of the painted surfaces and inner layers. As an example of this trend, in the framework of the interdisciplinary program “Picasso: Blue Project” recently carried out at the Museu Picasso in Barcelona, a selection of paintings of the Blue Period (1901–1904) collection were revisited through the eye of cutting-edge analytical techniques, including reflectance HSI. The present paper focuses on the painting “The Blue Glass” (La Copa Blava, 1903), which was examined by using a high-performance reflectance HSI instrumentation. Under its visible surface, this iconic oil on canvas featured a stratification of different compositions that perfectly witnessed the Picasso's creative process. The HSI data enabled an unprecedented visualisation of the artistic metamorphosis of this painting. The well-known Picasso's creative practice of repainting on his own canvas was brought to the light by means of HSI images, which offered unique materials for visualising the hidden layers of the painting, by disclosing subsequent and unknown earlier versions and the artistic evolution of the painting. The outcomes of the project were disseminated in an exhibition, that addressed the ambitious objective of sharing with the wide public scholarly advanced knowledge hitherto gained on the Picasso Blue Period. Within this context, reflectance HSI proved to be exceptionally effective, not only for addressing analytical tasks, but also for educational purposes.

Detection of early bruises on apples using hyperspectral reflectance imaging coupled with optimal wavelengths selection and improved watershed segmentation algorithm.

Bruises caused by mechanical collision during the harvesting and storage and transportation period are difficult to detect using traditional machine vision technologies because there is no obvious difference in appearance between bruised and sound tissues. As a result of its fast and non-destructive characteristics, hyperspectral imaging technology is a potential tool for non-destructive detection of fruit surface defects. In the present study, visible near infrared hyperspectral reflectance images of healthy apples and bruised apples at 6, 12 and 24 h were obtained. To reduce hyperspectral data dimension, optimal wavelength selection algorithms including principal component analysis (PCA) and band ratio methods were utilized to select the effective wavelengths and enhance the contrast between bruised and sound tissues. Then pseudo-color image transformation technology combining with improved watershed segmentation algorithm (IWSA) were employed to recognize the bruise spots. The result obtained showed that band ratio images obtained better detection performance than that of PCA. The G component derived from pseudo-color image of followed by IWSA obtained the best segmentation performance for bruise spots. Finally, a multispectral imaging system for the detection of bruised apple was developed to verify the effectiveness of the proposed two-band ratio algorithm, obtaining recognition rates of 93.3%, 92.2% and 92.5% for healthy, bruised and overall apples, respectively. The bruise detection algorithm proposed in the present study has potential to detect bruised apple in online practical applications and hyperspectral reflectance imaging offers a useful reference for the detection of surficial defects of fruit. © 2023 Society of Chemical Industry.

Measurement of rat and human tissue optical properties for improving the optical detection and visualization of peripheral nerves.

Peripheral nerve damage frequently occurs in challenging surgical cases resulting in high costs and morbidity. Various optical techniques have proven effective in detecting and visually enhancing nerves, demonstrating their translational potential for assisting in nerve-sparing medical procedures. However, there is limited data characterizing the optical properties of nerves in comparison to surrounding tissues, thus limiting the optimization of optical nerve detection systems. To address this gap, the absorption and scattering properties of rat and human nerve, muscle, fat, and tendon were determined from 352-2500 nm. The optical properties highlighted an ideal region in the shortwave infrared for detecting embedded nerves, which remains a significant challenge for optical approaches. A 1000-1700 nm hyperspectral diffuse reflectance imaging system was used to confirm these results and identify optimal wavelengths for nerve imaging contrast in an in vivo rat model. Optimal nerve visualization contrast was achieved using 1190/1100 nm ratiometric imaging and was sustained for nerves embedded under ≥600 µm of fat and muscle. Overall, the results provide valuable insights for optimizing the optical contrast of nerves, including those embedded in tissue, which could lead to improved surgical guidance and nerve-sparing outcomes.

Copper concentrate dual-band joint classification using reflectance hyperspectral images in the VIS-NIR and SWIR bands.

A study on the classification of copper concentrates relevant to the copper refining industry is performed by means of reflectance hyperspectral images in the visible and near infrared (VIS-NIR) bands (400-1000nm) and in the short-wave infrared (SWIR) (900-1700nm) band. A total of 82 copper concentrate samples were press compacted into 13-mm-diameter pellets, and their mineralogical composition was characterized via quantitative evaluation of minerals and scanning electron microscopy. The most representative minerals contained in these pellets are bornite, chalcopyrite, covelline, enargite, and pyrite. Three databases (VIS-NIR, SWIR, and VIS-NIR-SWIR) containing a collection of average reflectance spectra computed from 9×9p i x e l neighborhoods in each pellet hyperspectral image are compiled to train the classification models. The classification models tested in this work are a linear discriminant classifier and two non-linear classifiers, a quadratic discriminant classifier, and a fine K-nearest neighbor classifier (FKNNC). The results obtained show that the joint use of VIS-NIR and SWIR bands allows for the accurate classification of similar copper concentrates that contain only minor differences in their mineralogical composition. Specifically, among the three tested classification models, the FKNNC performs the best in terms of overall classification accuracy, achieving 93.4% accuracy in the test set when only VIS-NIR data are used to construct the classification model, up to 80.5% using only SWIR data, and up to 97.6% using both VIS-NIR and SWIR bands together.

Fine-scale mapping of physicochemical and microbial landscapes of the coral skeleton.

The coral skeleton harbours a diverse community of bacteria and microeukaryotes exposed to light, O2 and pH gradients, but how such physicochemical gradients affect the coral skeleton microbiome remains unclear. In this study, we employed chemical imaging of O2 and pH, hyperspectral reflectance imaging and spatially resolved taxonomic and inferred functional microbiome characterization to explore links between the skeleton microenvironment and microbiome in the reef-building corals Porites lutea and Paragoniastrea benhami. The physicochemical environment was more stable in the deep skeleton, and the diversity and evenness of the bacterial community increased with skeletal depth, suggesting that the microbiome was stratified along the physicochemical gradients. The bulk of the coral skeleton was in a low O2 habitat, whereas pH varied from pH 6-9 with depth. Physicochemical gradients of O2 and pH of the coral skeleton explained the β-diversity of the bacterial communities, and skeletal layers that showed O2 peaks had a higher relative abundance of endolithic algae, reflecting a link between the abiotic environment and the microbiome composition. Our study links the physicochemical, microbial and functional landscapes of the coral skeleton and provides new insights into the involvement of skeletal microbes in the coral holobiont metabolism.

HIDSAG: Hyperspectral Image Database for Supervised Analysis in Geometallurgy

Supervised analysis using spectral data requires a well-informed characterisation of the response variables and abundant spectral data points. The presented hyperspectral dataset comes from five sets of geometallurgical samples, each characterised by different methods. To provide the spectral data, all mineral samples were scanned with SPECIM VNIR and SWIR hyperspectral cameras. For each subset the following data are provided 1) hyperspectral reflectance images in the VNIR spectral range (400–1000 nm wavelength); 2) hyperspectral reflectance images in the SWIR spectral range (900–2500 nm wavelength); 3) hyperspectral reflectance images in the VNIR-SWIR range (merged to SWIR spatial resolution); 4) RGB images constructed from hyperspectral data using a Bilateral Filter based sensor fusion method; 5) response variables representing mineral sample characterisation results, provided as training and validation data. This dataset is intended for use in general regression and classification research and experiments. All subsets were validated using machine learning models with satisfactory results.

Hyperspectral reflectance imaging for nondestructive evaluation of root rot in Korean ginseng (Panax ginseng Meyer).

Root rot of Panax ginseng caused by Cylindrocarpon destructans, a soil-borne fungus is typically diagnosed by frequently checking the ginseng plants or by evaluating soil pathogens in a farm, which is a time- and cost-intensive process. Because this disease causes huge economic losses to ginseng farmers, it is important to develop reliable and non-destructive techniques for early disease detection. In this study, we developed a non-destructive method for the early detection of root rot. For this, we used crop phenotyping and analyzed biochemical information collected using the HSI technique. Soil infected with root rot was divided into sterilized and infected groups and seeded with 1-year-old ginseng plants. HSI data were collected four times during weeks 7-10 after sowing. The spectral data were analyzed and the main wavelengths were extracted using partial least squares discriminant analysis. The average model accuracy was 84% in the visible/near-infrared region (29 main wavelengths) and 95% in the short-wave infrared (19 main wavelengths). These results indicated that root rot caused a decrease in nutrient absorption, leading to a decline in photosynthetic activity and the levels of carotenoids, starch, and sucrose. Wavelengths related to phenolic compounds can also be utilized for the early prediction of root rot. The technique presented in this study can be used for the early and timely detection of root rot in ginseng in a non-destructive manner.

Classification of tree species based on hyperspectral reflectance images of stem bark

ABSTRACT Automatization of tree species identification in the field is crucial in improving forest-based bioeconomy, supporting forest management, and facilitating in situ data collection for remote sensing applications. However, tree species recognition has never been addressed with hyperspectral reflectance images of stem bark before. We investigated how stem bark texture differs between tree species using a hyperspectral camera set-up and gray level co-occurrence matrices and assessed the potential of using reflectance spectra and texture features of stem bark to identify tree species. The analyses were based on 200 hyperspectral reflectance data cubes (415–925 nm) representing ten tree species. There were subtle interspecific differences in bark texture. Using average spectral features in linear discriminant analysis classifier resulted in classification accuracy of 92–96.5%. Using spectral and texture features together resulted in accuracy of 93–97.5%. With a convolutional neural network, we obtained an accuracy of 94%. Our study showed that the spectral features of stem bark were robust for classifying tree species, but importantly, bark texture is beneficial when combined with spectral data. Our results suggest that in situ imaging spectroscopy is a promising sensor technology for developing accurate tree species identification applications to support remote sensing.

Identification of Chilling Injury in Kiwifruit Using Hyperspectral Structured-Illumination Reflectance Imaging System (SIRI) with Support Vector Machine (SVM) Modelling

Accurate detection of chilling injury in kiwifruit is challenging because the symptoms are mainly manifested in the interior. This work reports a method for detecting the chilling injury of ‘Hongyang’ kiwifruit to provide nondestructive discrimination. Kiwifruit samples with varying levels of chilling injury were analyzed by a hyperspectral structured-illumination reflectance imaging (SIRI) system. After demodulation, direct current (DC) and alternating current (AC) images with spatial frequencies of 30, 60, and 120 m−1 were obtained and labeled as F30, F60, and F120. Predictive models were developed to optimize the preprocessing and modeling methods. Prediction models established the results of DC and AC with different spatial frequencies and were compared. The autoscale-support vector machine (SVM) models were optimal for AC at different spatial frequencies, and the multiplicative scatter correction (MSC)-SVM model was optimal for DC. The combined features of F30, F60, and F120, as well as the spectral features of DC, had better accuracy for classifying chilling injury. The optimal model of hyperspectral SIRI system for detecting chilling injury was the F30 based on combined features, with calibration accuracy of 98.1% and prediction accuracy of 94.2%. This study has shown that structured illumination had higher accuracy than uniform illumination in predicting chilling injury. Further, this approach allows the identification of kiwifruit with chilling injury using a hyperspectral structured-illumination reflectance imaging system.