Spectral Unmixing Methods Research Articles

Precision crop mapping: within plant canopy discrimination of crop and soil using multi-sensor hyperspectral imagery

Leveraging diverse optomechanical and imaging technologies for precision agriculture applications is gaining attention in emerging economies. The precise spatial detection of plant objects in farms is crucial for optimizing plant-level nutrition and managing pests and diseases. High-resolution remote sensors mounted on drones have been increasingly deployed for large-scale crop mapping and field variability characterization. While field-level crop identification and crop-soil discrimination have been studied extensively, within-plant canopy discrimination of crop and soil has not been explored in real agricultural farms. The objectives of this study are: (i) adoption and assessment of spectral unmixing for discriminating crop and soil at within-plant canopy level, and (ii) generation of benchmark terrestrial and drone-based hyperspectral datasets for plant or sub-plant level discrimination using various spectral mixture modelling approaches and sources of endmembers. We acquired hyperspectral imagery of vegetable crops using a frame-based sensor mounted on a drone flying at different heights. Further, several linear, non-linear, and sparse-based spectral unmixing methods were used to discriminate plant and soil based on spectral signatures (endmembers) extracted from different spectral libraries prepared using in situ or field, ground-based, and drone-based hyperspectral imagery. The results, validated against pixel-to-pixel ground truth data, indicate an overall crop-soil discrimination accuracy of 99–100%, subject to a combination of endmember source and flying height. The influences of different endmember sources, spatial resolution as indicated by flying height, and inversion algorithms on the quality of estimated abundances are assessed from a verifiable and functionally relevant perspective. The generated hyperspectral datasets and ground truth data can be used for developing and testing new methods for sub-canopy level soil-crop discrimination in various agricultural applications of remote sensing.

Distribution-informed and wavelength-flexible data-driven photoacoustic oximetry.

Photoacoustic imaging (PAI) promises to measure spatially resolved blood oxygen saturation but suffers from a lack of accurate and robust spectral unmixing methods to deliver on this promise. Accurate blood oxygenation estimation could have important clinical applications from cancer detection to quantifying inflammation. We address the inflexibility of existing data-driven methods for estimating blood oxygenation in PAI by introducing a recurrent neural network architecture. We created 25 simulated training dataset variations to assess neural network performance. We used a long short-term memory network to implement a wavelength-flexible network architecture and proposed the Jensen-Shannon divergence to predict the most suitable training dataset. The network architecture can flexibly handle the input wavelengths and outperforms linear unmixing and the previously proposed learned spectral decoloring method. Small changes in the training data significantly affect the accuracy of our method, but we find that the Jensen-Shannon divergence correlates with the estimation error and is thus suitable for predicting the most appropriate training datasets for any given application. A flexible data-driven network architecture combined with the Jensen-Shannon divergence to predict the best training data set provides a promising direction that might enable robust data-driven photoacoustic oximetry for clinical use cases.

Using satellite image data to identify rice varieties through linear spectral unmixing method (case study: Karangjati Sub District, Ngawi Regency)

<p>Remote sensing technology has increasingly emerged as a potent tool for precision agriculture, particularly in facilitating the mapping and monitoring of crops on a large scale. An application of this technology is the identification of different types of rice by analyzing the pixels acquired in satellite images. Regrettably, the pixels in the image have been mixed from different recorded items. Therefore, they have the potential to influence the outcome of the identification. An effective approach to addressing this problem is to employ the linear spectral unmixing (LSU) technique. The LSU approach quantifies the ratio of pure objects in every pixel of an image by utilizing the spectral value associated with the endmember of the rice variety. The investigation was carried out in the Karangjati District during the generative stage (70 ± DAP) of the rice planting season. The data indicates that the dominant variety is Inpari 32 HDB. The data validation tests, which involved the use of a confusion matrix and Kappa analysis, resulted in an overall accuracy rate of 85.48% and a Kappa analysis score of 70.6%.<strong></strong></p>

Estimation of Rice Plant Coverage Using Sentinel-2 Based on UAV-Observed Data

Vegetation coverage is a crucial parameter in agriculture, as it offers essential insight into crop growth and health conditions. The spatial resolution of spaceborne sensors is limited, hindering the precise measurement of vegetation coverage. Consequently, fine-resolution ground observation data are indispensable for establishing correlations between remotely sensed reflectance and plant coverage. We estimated rice plant coverage per pixel using time-series Sentinel-2 Multispectral Instrument (MSI) data, enabling the monitoring of rice growth conditions over a wide area. Coverage was calculated using unmanned aerial vehicle (UAV) data with a spatial resolution of 3 cm with the spectral unmixing method. Coverage maps were generated every 2–3 weeks throughout the rice-growing season. Subsequently, crop growth was estimated at 10 m resolution through multiple linear regression utilizing Sentinel-2 MSI reflectance data and coverage maps. In this process, a geometric registration of MSI and UAV data was conducted to improve their spatial agreement. The coefficients of determination (R2) of the multiple linear regression models were 0.92 and 0.94 for the Level-1C and Level-2A products of Sentinel-2 MSI, respectively. The root mean square errors of estimated rice plant coverage were 10.77% and 9.34%, respectively. This study highlights the promise of satellite time-series models for accurate estimation of rice plant coverage.

A phenology-based linear spectral unmixing method for rice varieties identification from Landsat 8 Satellite Image in Ngawi District, East Java, Indonesia

Abstract. Cipta IM, Jaelani LM, Sanjaya H. 2024. A phenology-based linear spectral unmixing method for rice varieties identification from Landsat 8 Satellite Image in Ngawi District, East Java, Indonesia. Biodiversitas 25: 1691-1702. Ngawi District is a prominent rice-producing district in East Java, Indonesia. Based on the BPS data from 2021, Ngawi District yielded a total of 786.49 thousand tons of paddy. Multiple strategies to enhance rice production include enhancing rice productivity, increasing paddy fields, and using effective land management techniques. Utilizing a superior breed of seeds can augment productivity. Remote sensing can be utilized to obtain such data, one of which is the Linear Spectral Unmixing (LSU) technique. The LSU approach is used to determine the proportion of a pure object's presence in a pixel. The Inpari 32 rice variety is the predominant type identified from the analysis of the processing outcomes utilizing satellite imagery. In addition, validation was performed by utilizing many sample locations distributed around Ploso Lor and Rejuno Villages, resulting in consistent findings between the data analysis and the actual conditions observed in the field. The RMSE calculation was conducted on both sets of processing results, and the most optimal RMSE value was achieved when processing was performed using Landsat 8 image data, yielding an RMSE value of roughly ±1.2%. The accuracy of the model was evaluated using a confusion matrix, which yielded an overall accuracy rate of 86.67%.

Classification of Plenodomus lingam and Plenodomus biglobosus in Co-Occurring Samples Using Reflectance Spectroscopy

Under natural conditions, mixed infections are often observed when two or more species of plant pathogens are present on one host. Thus, the detection and characterization of co-occurring pest species is a challenge of great importance. In this study, we focused on the development of a spectral unmixing method for the discrimination of two fungi species, Plenodomus lingam and Plenodomus biglobosus, the pathogens of oilseed rape. Over 24 days, spectral reflectance measurements from Petri dishes inoculated with fungi were conducted. Four experimental combinations were used: the first two were pure fungal samples, while the other two were co-occurring fungal samples. The results of the study show the possibility of distinguishing, based on spectral characteristics, between P. lingam and P. biglobosus not only in pure but also in co-occurring samples. We observed the changes in the reflectance of electromagnetic radiation from the tested fungi over time and a strong correlation between the reflectance and changes in the areas of the mycelia on the Petri dishes. Moreover, the wavelengths most useful for spectral classification of the tested fungal mycelia were selected. Finally, a spectral unmixing model was proposed, which enables the estimation of the areas of two pathogens in co-occurring samples based on the spectral characteristics of the entire plate with an error smaller than 0.2. To our knowledge, the present study is the first report examining the use of reflectance spectroscopy methods for classifying pathogens on the same Petri dish. The study results indicate the feasibility of reflectance spectroscopy as a nondestructive sampling method for plant pathogen detection and classification.

Timely monitoring of soil water-salt dynamics within cropland by hybrid spectral unmixing and machine learning models

Soil salinization and water scarcity are main restrictive factors for irrigated agriculture development in arid regions. Knowing dynamics of soil water and salt content is an important antecedent in remediating salinized soils and optimizing irrigation management. Previous studies mostly used remote sensing technologies to individually monitor water or salt content dynamics in agricultural areas. Their ability to asses different levels of crop water and salt management has been less explored. Therefore, how to extract effective diagnostic features from remote sensing images derived spectral information is crucial for accurately estimating soil water and salt content. In this study, Linear spectral unmixing method (LSU) was used to obtain the contribution of soil water and salt to each band spectrum (abundance), and endmember spectra from Sentinel-2 images. Calculating spectral indices and selecting optimal spectal combination were individually based on soil water and salt endmember spectra. The estimation models were constructed using six machine learning algorithms: BP Neural Network (BPNN), Support Vector Regression (SVR), Partial Least Squares Regression (PLSR), Random Forest Regression (RFR), Gradient Boost Regression Tree (GBRT), and eXtreme Gradient Boosting tree (XGBoost). The results showed that the spectral indices calculated from endmember spectra were able to effectively characterize the response of crop spectral properties to soil water and salt, which circumvent spectral ambiguity induced by water-salt mixing. NDRE spectral index was a reliable indicator for estimating water and salt content, with determination coefficients (R2) being 0.55 and 0.57, respectively. Compared to other models, LSU-XGBoost model achieved the best performance. This model properly reflected the process of soil water-salt dynamics in farmland during crop growth period. This study provided new methods and ideas for soil water-salt estimation in dry irrigated agricultural areas, and provided decision support for governance of salinized land and optimal management of irrigation.

Enhancing Solar-Induced Fluorescence Interpretation: Quantifying Fractional Sunlit Vegetation Cover Using Linear Spectral Unmixing

Solar-induced chlorophyll fluorescence (SIF) is closely related to plant photosynthetic activity and has been used in different studies as a proxy for vegetation health status. However, in order to use SIF as a relevant indicator of plant physiological stress, it is necessary to accurately quantify the amount of light absorbed by the photosynthetic plant pigments, called the absorbed photosynthetically active radiation (APAR). The ratio between fluorescence emission and light absorption (i.e., SIF and APAR) is known as the fluorescence quantum efficiency (FQE). In this work, simultaneous measurements of SIF and reflected radiance were performed both at the leaf and canopy levels for Salvia farinacea and Datura stramonium plants. With the aim of disentangling the proportion of sunlit and shaded absorbed PAR, an ad hoc experimental setup was designed to provide a wide range of fraction vegetation cover (FVC) canopy settings. A linear spectral unmixing method was proposed to estimate the contribution of soil, sunlit, and shaded vegetation from the total reflectance spectrum measured at the canopy level. Later, the retrieved sunlit FVC (FVCsunlit) was used to estimate the (dominant) green APAR flux, and this was combined with the integral of the spectrally resolved fluorescence to calculate the FQE. The results of this study demonstrated that under no-stress conditions and independently of the FVC, similar FQE values were observed when SIF was properly normalised by the green APAR flux. The results obtained showed that the reflectance spectra retrieved using a linear unmixing method had a maximum RMSE of less than 0.03 along the spectrum. The FVCsunlit evaluation showed an RMSE of 14% with an R2 of 0.84. Moreover, the FQE values obtained at the top of the canopy (TOC) were found statistically comparable to the reference values at the leaf level. These results support further efforts to improve the interpretation of fluorescence based on field spectroscopy and the further upscaling to imaging spectroscopy at airborne and satellite levels.

AutoUnmix: an autoencoder-based spectral unmixing method for multi-color fluorescence microscopy imaging.

Multiplexed fluorescence microscopy imaging is widely used in biomedical applications. However, simultaneous imaging of multiple fluorophores can result in spectral leaks and overlapping, which greatly degrades image quality and subsequent analysis. Existing popular spectral unmixing methods are mainly based on computational intensive linear models, and the performance is heavily dependent on the reference spectra, which may greatly preclude its further applications. In this paper, we propose a deep learning-based blindly spectral unmixing method, termed AutoUnmix, to imitate the physical spectral mixing process. A transfer learning framework is further devised to allow our AutoUnmix to adapt to a variety of imaging systems without retraining the network. Our proposed method has demonstrated real-time unmixing capabilities, surpassing existing methods by up to 100-fold in terms of unmixing speed. We further validate the reconstruction performance on both synthetic datasets and biological samples. The unmixing results of AutoUnmix achieve the highest SSIM of 0.99 in both three- and four-color imaging, with nearly up to 20% higher than other popular unmixing methods. For experiments where spectral profiles and morphology are akin to simulated data, our method realizes the quantitative performance demonstrated above. Due to the desirable property of data independency and superior blind unmixing performance, we believe AutoUnmix is a powerful tool for studying the interaction process of different organelles labeled by multiple fluorophores.

Regression-based surface water fraction mapping using a synthetic spectral library for monitoring small water bodies

ABSTRACT Small water bodies (SWBs), such as ponds and on-farm reservoirs, are a key part of the hydrological system and play important roles in diverse domains from agriculture to conservation. The monitoring of SWBs has been greatly facilitated by medium-spatial-resolution satellite images, but the monitoring accuracy is considerably affected by the mixed-pixel problem. Although various spectral unmixing methods have been applied to map sub-pixel surface water fractions for large water bodies, such as lakes and reservoirs, it is challenging to map SWBs that are small in size relative to the image pixel and have dissimilar spectral properties. In this study, a novel regression-based surface water fraction mapping method (RSWFM) using a random forest and a synthetic spectral library is proposed for mapping 10 m spatial resolution surface water fractions from Sentinel-2 imagery. The RSWFM inputs a few endmembers of water, vegetation, impervious surfaces, and soil to simulate a spectral library, and considers spectral variations in endmembers for different SWBs. Additionally, RSWFM applies noise-based data augmentation on pure endmembers to overcome the limitation often arising from the use of a small set of pure spectra in training the regression model. RSWFM was assessed in ten study sites and compared with the fully constrained least squares (FCLS) linear spectral mixture analysis, multiple endmember spectral mixture analysis (MESMA), and the nonlinear random forest (RF) regression without data-augmentation. The results showed that RSWFM decreases the water fraction mapping errors by ~ 30%, ~15%, and ~ 11% in root mean square error compared with the linear FCLS, MESMA unmixings, and the nonlinear RF regression without data-augmentation respectively. RSWFM has an accuracy of approximately 0.85 in R2 in estimating the area of SWBs smaller than 1 ha.

Endmember variability based abundance estimation of red and black soil over sparsely vegetated area using AVIRIS-NG hyperspectral image

Visible-near-and-short-wave-infrared hyperspectral images (HSI) have proven helpful for mapping the soil types over bare soils pixels. However, the accuracy of the traditional pixel-based classification methods decreases due to the spectral mixing between the significant agricultural features such as vegetation, soil and crop residue. In this context, spectral unmixing algorithms are handy for estimating the sub-pixel abundances of soil over sparsely vegetated areas. Most of the spectral unmixing methods focus on analyzing the HSI by considering the pixels as independent entities. However, in the HSIs of agricultural fields, the endmembers are spatially and temporally variable. It is also known that there exists a strong spatial correlation among neighbourhood pixels over agricultural fields. Therefore, both endmember variability and the neighbourhood spatial-spectral information are critical to estimate soil abundance accurately. Here we address the issue of estimating the abundances of two major soil types with spectral variability. The proposed approach combines the endmember bundle extraction with the spectral-spatial weighted unmixing approach (SSWU-SV). Experimental analysis has been carried out on the airborne HSI acquired by airborne-visible-and-infrared-imaging- spectrometer-next-generation (AVIRIS-NG) sensor. The quantitative analysis reveals that the proposed method consistently achieves a better unmixing performance than the traditional linear mixing model in terms of spectral angle mapper (SAM), and root-mean-square error (RMSE). Our results also indicate the saturation of normalized difference vegetation index (NDVI) at high vegetative fractions obtained from SSWU-SV.

Implementation of orthogonal subspace projection on ASTER data for exploring phosphorite -a study in Chhatarpur, Madhya Pradesh, India

In this study, we have identified surface exposures of phosphorite using visible near infrared (VNIR) and shortwave infrared (SWIR) bands of Advanced space borne thermal emission and reflection radiometer (ASTER) data. We analysed the spectral contrast of dolomite and phosphorite samples collected from the ground and also from the pixels of the ASTER image composite. We observed the consistent spectral contrast between the samples of dolomite and phosphorite and used the same for delineating phosphorite within the dolomite and associated carbonate rocks. In this regard, we first delineated dolomite and associated carbonate rocks using feature oriented principal component of the iron rich carbonate rocks. Subsequently, we implemented partial spectral unmixing method, known as orthogonal subspace projection on ASTER VNIR and SWIR bands to delineate phosphorite within dolomite. We could trace a few new occurrences of phosphorite in the study area beyond the known occurrence at Hirapur open cast mining area. We have validated the identified new areas during field survey using a rapid colorimetric test for phosphate to ascertain the exploration importance of these targets. We further estimated P2O5 grade content of the phosphorite samples in these sites and found few of them have very high P2O5 content.

Measurement of chemical penetration in skin using Stimulated Raman scattering microscopy and multivariate curve resolution - alternating least squares

The mechanistic understanding of skin penetration underpins the design, efficacy and risk assessment of many high-value products including functional personal care products, topical and transdermal drugs. Stimulated Raman scattering (SRS) microscopy, a label free chemical imaging tool, combines molecular spectroscopy with submicron spatial information to map the distribution of chemicals as they penetrate the skin. However, the quantification of penetration is hampered by significant interference from Raman signals of skin constituents. This study reports a method for disentangling exogeneous contributions and measuring their permeation profile through human skin combining SRS measurements with chemometrics. We investigated the spectral decomposition capability of multivariate curve resolution – alternating least squares (MCR–ALS) using hyperspectral SRS images of skin dosed with 4-cyanophenol. By performing MCR-ALS on the fingerprint region spectral data, the distribution of 4-cyanophenol in skin was estimated in an attempt to quantify the amount permeated at different depths. The reconstructed distribution was compared with the experimental mapping of CN, a strong vibrational peak in 4-cyanophenol where the skin is spectroscopically silent. The similarity between MCR-ALS resolved and experimental distribution in skin dosed for 4 h was 0.79 which improved to 0.91 for skin dosed for 1 h. The correlation was observed to be lower for deeper layers of skin where SRS signal intensity is low which is an indication of low sensitivity of SRS. This work is the first demonstration, to the best of our knowledge, of combining SRS imaging technique with spectral unmixing methods for direct observation and mapping of the chemical penetration and distribution in biological tissues.

Blind spectral unmixing for characterization of plaque composition based on multispectral photoacoustic imaging

To improve the assessment of carotid plaque vulnerability, a comprehensive characterization of their composition is paramount. Multispectral photoacoustic imaging (MSPAI) can provide plaque composition based on their absorption spectra. However, although various spectral unmixing methods have been developed to characterize different tissue constituents, plaque analysis remains a challenge since its composition is highly complex and diverse. In this study, we employed an adapted piecewise convex multiple-model endmember detection method to identify carotid plaque constituents. Additionally, we explore the selection of the imaging wavelengths in linear models by conditioning the coefficient matrix and its synergy with our unmixing approach. We verified our method using plaque mimicking phantoms and performed ex-vivo MSPAI on carotid endarterectomy samples in a spectral range from 500 to 1300 nm to identify the main spectral features of plaque materials for vulnerability assessment. After imaging, the samples were processed for histological analysis to validate the photoacoustic decomposition. Results show that our approach can perform spectral unmixing and classification of highly heterogeneous biological samples without requiring an extensive fluence correction, enabling the identification of relevant components to assess plaque vulnerability.

Landsat, MODIS, and VIIRS snow cover mapping algorithm performance as validated by airborne lidar datasets

Abstract. Snow cover mapping algorithms utilizing multispectral satellite data at various spatial resolutions are available, each treating subpixel variation differently. Past evaluations of snow mapping accuracy typically relied on satellite data collected at a higher spatial resolution than the data in question. However, these optical data cannot characterize snow cover mapping performance under forest canopies or at the meter scale. Here, we use 3 m spatial resolution snow depth maps collected on 116 d by an aerial laser scanner to validate band ratio and spectral-mixture snow cover mapping algorithms. Such a comprehensive evaluation of sub-canopy snow mapping performance has not been undertaken previously. The following standard (produced operationally by an agency) products are evaluated: NASA gap-filled Moderate Resolution Imaging Spectroradiometer (MODIS) MOD10A1F, NASA gap-filled Visible Infrared Imaging Radiometer Suite (VIIRS) VNP10A1F, and United States Geological Survey (USGS) Landsat 8 Level-3 Fractional Snow Covered Area. Two spectral-unmixing approaches are also evaluated: Snow-Covered Area and Grain Size (SCAG) and Snow Property Inversion from Remote Sensing (SPIReS), both of which are gap-filled MODIS products and are also run on Landsat 8. We assess subpixel snow mapping performance while considering the fractional snow-covered area (fSCA), canopy cover, sensor zenith angle, and other variables within six global seasonal snow classes. Metrics are calculated at the pixel and basin scales, including the root-mean-square error (RMSE), bias, and F statistic (a detection measure). The newer MOD10A1F Version 61 and VNP10A1F Version 1 product biases (− 7.1 %, −9.5 %) improve significantly when linear equations developed for older products are applied (2.8 %, −2.7 %) to convert band ratios to fSCA. The F statistics are unchanged (94.4 %, 93.1 %) and the VNP10A1F RMSE improves (18.6 % to 15.7 %), while the MOD10A1F RMSE worsens (12.7 % to 13.7 %). Consistent with previous studies, spectral-unmixing approaches (SCAG, SPIReS) show lower biases (−0.1 %, −0.1 %) and RMSE (12.1 %, 12.0 %), with higher F statistics (95.6 %, 96.1 %) relative to the band ratio approaches for MODIS. Landsat 8 products are all spectral-mixture methods with low biases (−0.4 % to 0.3 %), low RMSE (11.4 % to 15.8 %), and high F statistics (97.3 % to 99.1 %). Spectral-unmixing methods can improve snow cover mapping at the global scale.

Spectral unmixing method considering endmember variability of vegetation

Spectral unmixing method considering endmember variability of vegetation

Urban Land-Cover Changes in Major Cities in China from 1990 to 2015.

The accelerated urbanization process in China has led to land-cover changes, triggering a series of environmental issues as one of the major drivers of global change. We studied the land-cover changes in the built-up areas of 50 major cities in China from 1990 to 2015 with Landsat data combined with spectral unmixing methods and decision tree classification. The overall accuracy of urban land-cover type products with 30 m resolution was obtained as 84%, which includes impervious surfaces, bare soil, vegetation, and water bodies. Based on these land-cover type products, the results show that the urbanization of major cities in China manifests itself as a steep expansion of impervious surfaces (+32.91%) and vegetation (+36.93%), while the proportion of bare soil (-68.64%) and water bodies (-1.20%) decreases. The increase in vegetation indicates an increasing emphasis on greening during urbanization, which is especially vital for the sustainability of urban ecosystems. Increasing economic standards and population sizes are significantly correlated with impervious surface expansion and may be the main drivers of urbanization. Nationwide, there is a decreasing trend of shape complexity among different large cities, which indicates that landscape shapes will gradually become regular when cities grow to a certain level. Greenspace areas in the cities increased significantly during 1990-2015 and became more fragmented and tended to disperse across cities. These changes reflect the government's efforts to enhance urban ecosystem functions to serve the rapidly increasing urban population in China over the past three decades.

Autofluorescence prediction model for fluorescence unmixing and age determination.

Flow cytometry is a powerful tool for identifying and quantifying various cell markers, such as viability, vitality, and individual cell age, at single-cell stages. However, cell autofluorescence and marker fluorophore signals overlap at low fluorescence intensities. Thus, these signals must be unmixed before determining the age fraction. A comparison was made between principal component regression (PCR) and random forest (RF) to predict autofluorescence signals of Saccharomyces pastorianus var. carlsbergensis in a flow cytometer. RF provided better prediction results than the PCR and was therefore determined to be better suited for unmixing signals. In the subsequent application for unmixing the autofluorescence signal from the marker fluorophore signal, the Gaussian mixture analysis based on RF was in better agreement with the microscopy-determined replicative age distribution than the PCR-based method. The proposed approach of single-laser spectral unmixing and subsequent Gaussian mixture analysis showed that the microscopy data was consistent with the unmixed fluorescence spectra. The demonstrated approach enables fast and reliable unmixing of flow cytometric spectral data using a single-laser spectral unmixing method. This analysis method enables age determination of cells in industrial processes. This age determination allows for quantifying the yeast cell's age fractions, providing a detailed view of age-related changes. Additionally, the bud scar labeling technique can be used to determine age-related changes in Pichia pastoris yeast for biotechnological applications or recombinant protein expression.

Nonlinear Unmixing via Deep Autoencoder Networks for Generalized Bilinear Model

Hyperspectral unmixing decomposes the observed mixed spectra into a collection of constituent pure material signatures and the associated fractional abundances. Because of the universal modeling ability of neural networks, deep learning (DL) techniques are gaining prominence in solving hyperspectral analysis tasks. The autoencoder (AE) network has been extensively investigated in linear blind source unmixing. However, the linear mixing model (LMM) may fail to provide good unmixing performance when the nonlinear mixing effects are nonnegligible in complex scenarios. Considering the limitations of LMM, we propose an unsupervised nonlinear spectral unmixing method, based on autoencoder architecture. Firstly, a deep neural network is employed as the encoder to extract the low-dimension feature of the mixed pixel. Then, the generalized bilinear model (GBM) is used to design the decoder, which has a linear mixing part and a nonlinear mixing one. The coefficient of the bilinear mixing part can be adjusted by a set of learnable parameters, which makes the method perform well on both nonlinear and linear data. Finally, some regular terms are imposed on the loss function and an alternating update strategy is utilized to train the network. Experimental results on synthetic and real datasets verify the effectiveness of the proposed model and show very competitive performance compared with several existing algorithms.

Identification of Paddy Varieties from Landsat 8 Satellite Image Data Using Spectral Unmixing Method in Indramayu Regency, Indonesia

Indramayu Regency is the highest rice producer in West Java province, Indonesia. According to the Central Statistics Agency (BPS), in 2021, rice production in 2020 reached 1,365,435.39 tons of GKG (milled dry grain). Technological developments in the food sector produce various kinds of premium quality rice and rice varieties resistant to climate change, such as Ciherang, Inpari 32 HDB and IR 64. The regular monitoring of specific rice varieties over large areas effectively maintains the quality and quantity of rice production. This study used remote sensing data to monitor rice conditions and distribution based on the spectral unmixing method. The spectral unmixing method was used to identify the percentage of the presence of a pure object in a pixel. The results obtained in this study were images of the endmember fractions of rice varieties and areas of dominant rice varieties used in the Indramayu district. The dominant variety detected with the processing results was the Inpari 32 HDB variety, with an area of 30,738.64 hectares. In comparison, varieties other than Inpari 32 HDB were also detected in several areas in the Indramayu district, with an area of 12,192.68 hectares.