Band Selection Research Articles

Superpixel guided spectral-spatial feature extraction and weighted feature fusion for hyperspectral image classification with limited training samples

Deep learning is a double-edged sword. The powerful feature learning ability of deep models can effectively improve classification accuracy. Still, when the training samples for each class are limited, it will not only face the problem of overfitting but also significantly affect the classification result. Aiming at this critical problem, we propose a novel model of spectral-spatial feature extraction and weighted fusion guided by superpixels. It aims to thoroughly “squeeze” and utilize the untapped spectral and spatial features contained in hyperspectral images from multiple angles and stages. Firstly, with the guidance of superpixels, we represent the hyperspectral image in the form of latent features and use the multi-band priority criterion to select the final discriminant features. Secondly, we design a pixel-based CNN and a two-scale superpixel-based GCN classification framework for weighted feature fusion. Compared with several excellent band selection methods, the superb performance of our feature extraction module is verified. In addition, under the condition of only five training samples for each class, we conducted comparative experiments with several of the state-of-the-art classification methods and verified the excellent performance of our method on three widely used data sets.

Hyperspectral Band Selection Method Based on Global Partition Clustering

Band selection is an important step in the dimensionality reduction processing of hyperspectral images and is highly important for eliminating redundant spectral information and reducing computational costs. In recent years, band selection methods based on ordered partition have been widely used in the dimensionality reduction processing of hyperspectral images. However, most of the methods use coarse and fine partition for band subspace partition, so that the partition results are affected by equal interval partition. Furthermore, existing methods usually select a representative band in each band subspace but do not consider the relationship between the output bands in the selection process, resulting in a certain degree of redundancy between the output bands. To solve the above problems, we propose a band selection method based on global partition clustering that contains band subspace partition and band selection. The band subspace partition is based on coarse and fine partition and the similarity-based ranking-structural similarity method, which divides the band space into band subspaces according to the relationship between the number of bands in the hyperspectral image and the number of selected bands. Band selection is based on the sequential forward selection method, which iteratively selects one band as the output band in each band subspace. The proposed method makes two main contributions. Firstly, this method avoids the negative effect of equal interval partition on the results. Secondly, this method fully considers the relationship between the selected bands and the bands to be selected. The effectiveness of the proposed method is demonstrated via ablation and comparison experiments on three publicly available datasets. Comparison experiments show that the classification accuracy of the proposed method can exceed the accuracy of the comparison methods.

Automating band selection for hyperspectral indices: bridging AVIRIS-NG and Sentinel-2 satellite data for earth science applications

Hyperspectral data from the Airborne Visible and Infra-Red Imaging Spectrometer – Next-Generation (AVIRIS-NG) offers transformative potential for Earth science research, enabling detailed analysis of land surface processes, resource monitoring, and environmental dynamics. This study presents an automated methodology to optimize the selection of AVIRIS spectral bands, improving the computation of indices critical to Earth science applications. By leveraging multiple hyperspectral bands, the approach enhances the accuracy of indices used to monitor water resources, vegetation health, urban expansion, and built-up areas. The methodology involves calculating indices from all possible AVIRIS band combinations, evaluating their root mean squared error (RMSE) against Sentinel-2 indices, reducing RMSE skewness, and selecting bands with minimal deviation for specific Land Use Land Cover (LULC) categories. The process is automated and employs parallel processing with Python, significantly reducing execution time and enabling scalability for large geospatial datasets. Key indices, including the Normalized Difference Water Index (NDWI), Normalized Difference Red Edge (NDRE), and Normalized Difference Built-up Index (NDBI), Green Normalized Difference Vegetation Index (GNDVI) were validated using the proposed methodology. Results demonstrate the potential of hyperspectral data to outperform traditional single-band approaches, providing more precise and reliable assessments.

Improving Soil Heavy Metal Lead Inversion Through Combined Band Selection Methods: A Case Study in Gejiu City, China

Hyperspectral technology has become increasingly important in monitoring soil heavy metal pollution, yet hyperspectral data often contain substantial band redundancy, and band selection methods are typically limited to single algorithms or simple combinations. Multi-algorithm combinations for band selection remain underutilized. To address this gap, this study, conducted in Gejiu, Yunnan Province, China, proposes a multi-algorithm band selection method to enable the rapid prediction of lead (Pb) contamination levels in soil. To construct a preliminary Pb content prediction model, the initial selection of spectral bands utilized methods including CARS (Competitive Adaptive Reweighted Sampling), GA (Genetic Algorithm), MI (mutual information), SPA (Successive Projections Algorithm), and WOA (Whale Optimization Algorithm). The results indicated that WOA achieved the highest modeling accuracy. Building on this, a combined WOA-based band selection method was developed, including combinations such as WOA-CARS, WOA-GA, WOA-MI, and WOA-SPA, with multi-level band optimization further refined by MI (e.g., WOA-GA-MI, WOA-CARS-MI, WOA-SPA-MI). The results showed that the WOA-GA-MI model exhibited optimal performance, achieving an average R2 of 0.75, with improvements of 0.32, 0.11, and 0.02 over the full-spectrum model, the WOA-selected spectral model, and the WOA-GA model, respectively. Additionally, spectral response analysis identified 22 common bands essential for Pb content inversion. The proposed multi-level combined model not only significantly enhances prediction accuracy but also provides new insights into optimizing hyperspectral band selection, serving as a valuable scientific foundation for assessing soil heavy metal contamination.

Stepwise discriminant analysis based optimal frequency band selection and ensemble learning for same limb MI recognition

Stepwise discriminant analysis based optimal frequency band selection and ensemble learning for same limb MI recognition

Nature-based metaheuristic optimisation techniques for band selection in drone-based hyperspectral images for plant-level crop classification

Nature-based metaheuristic optimisation techniques for band selection in drone-based hyperspectral images for plant-level crop classification

Hyperspectral Band Selection with Unique Pixel Extraction and Adaptive Neighbor Clustering

Band selection is an effective way to reduce redundant information, while preserving the physical properties of hyperspectral images (HSI). However, most band selection methods merely consider the relevance and separability between pairs of bands and ignore those for different ground objects. To solve these issues, we propose a Unique Pixel extraction and Adaptive Neighbor Clustering (UPANC) band selection method in this theoretical study. First, in consideration of the characteristics of HSI data and tasks, unique pixels are obtained with a low-rank representation, where the importance of bands is analyzed from both spectral and spatial perspectives. Second, an adaptive neighbor clustering method is designed based on the unique pixels, which groups bands into several clusters through optimizing the graph structure under label smoothness. With support vector machines (SVM) as the classifier, the UPANC method achieved good performance, where the overall accuracy scores were 89.05%, 82.62%, and 92.07% on the Houston, IndianPines, and Pavia University datasets, respectively. The experimental results illustrated the advantages of the UPANC method, which could select optimal bands to enhance the performance in land cover observation.

Optimal Imaging Band Selection for Laser-Vision System Based on Welding Arc Spectrum Analysis

Optimal Imaging Band Selection for Laser-Vision System Based on Welding Arc Spectrum Analysis

Identifying Optimal Variables to Predict Soil Organic Carbon in Sandy, Saline, and Black Soil Regions: Remote Sensing, Terrain, or Climate Factors?

Environmental variables have a substantial effect on the reliability of soil organic carbon (SOC) mapping. However, it is still challenging to identify which environmental variables are effective in cropland SOC prediction in sandy, saline, and black soil regions. To address this issue, we used the principal component analysis (PCA) method for the feature selection of bands, spectral indexes, and terrain factors for each region. Based on the selection feature, we used global RF and local RF for SOC prediction for these three regions. Our results indicated that (1) climate factors, particularly mean annual precipitation and mean annual temperature, were the most effective predictors in SOC mapping across sandy, saline, and black soil regions, as indicated by their significant contribution to RF model performance (R2 > 0.63); (2) followed by climate factors, the Transformed Vegetation Index (TVI) was consistently identified as the most influential variable for SOC prediction among spectral indexes in all three regions; (3) a local regression method based on RF models showed good performance compared to a global model; (4) desertification and salinization were the main reasons for the spatial differences in AH and DM&LD, respectively. The SOC of HL in black soil regions was consistent with the climate change trend because of the latitude difference. This study provides valuable information for constructing a more precise soil prediction strategy for cultivated land in sandy, saline, and black soil regions.

Band Selection Algorithm Based on Multi-Feature and Affinity Propagation Clustering

Band Selection Algorithm Based on Multi-Feature and Affinity Propagation Clustering

Plastic waste identification based on multimodal feature selection and cross-modal Swin Transformer

The classification and recycling of municipal solid waste (MSW) are strategies for resource conservation and pollution prevention, with plastic waste identification being an essential component of waste sorting. Multimodal detection of solid waste has increasingly replaced single-modal methods constrained by limited informational capacity. However, existing hyperspectral feature selection algorithms and multimodal identification methods have yet to leverage cross-modal information exhaustively. Therefore, two RGB-hyperspectral image (RGB-HSI) multimodal instance segmentation datasets were constructed to support research in plastic waste sorting. A feature band selection algorithm based on the Activation Weight function was proposed to automatically select influential hyperspectral bands from multimodal data, thereby reducing the burden of data acquisition, transmission, and inference. Furthermore, the multimodal Selective Feature Network (SFNet) was introduced to balance information across various modalities and stages. Moreover, the Correlation Swin Transformer Block was proposed, specifically crafted to fuse cross-modal mutual information, which can be synergistically employed with SFNet to enhance multimodal recognition capabilities further. Experimental results show that the Activation Weight band selection function can select the most effective feature bands. At the same time, the Correlation SF-Swin Transformer achieved the highest F1-scores of 97.85% and 97.37% in the two plastic waste object detection experiments, respectively. The source code and final models are available at https://github.com/Bazenr/Correlation-SFSwin, and the dataset can be accessed at https://www.kaggle.com/datasets/bazenr/rgb-hsi-rgb-nir-municipal-solid-waste.

Optimal band selection and transfer in drone-based hyperspectral images for plant-level vegetable crops identification using statistical-swarm intelligence (SSI) hybrid algorithms

Optimal band selection and transfer in drone-based hyperspectral images for plant-level vegetable crops identification using statistical-swarm intelligence (SSI) hybrid algorithms

Non-destructive detection of milk nutritional components based on hyperspectral imaging.

As consumers increasingly prioritize food safety and nutritional value, the dairy industry faces a pressing need for rapid and accurate methods to detect essential nutritional components in milk, such as fat, protein, and lactose. Hyperspectral imaging (HSI) technology, known for its non-destructive, fast, and precise nature, shows great promise in food quality assessment. However, the high dimensionality of HSI data poses challenges for effective band selection and model optimization. Additionally, prior studies primarily focus on predicting single nutritional components without addressing simultaneous multi-component detection. To overcome these challenges, this study presents a comprehensive approach that integrates moving average smoothing and first derivative (MA-FD) preprocessing, the improved coati optimization algorithm (ICOA), and the CatBoost model for multi-target regression. ICOA incorporates the good point set strategy, dynamic opposition-based learning, and the golden sine algorithm, which significantly enhance its global search capability and convergence speed in band selection. Combined with CatBoost's multi-target prediction capability, this method enables accurate detection of fat, protein, and lactose levels in milk. Experimental results demonstrate high prediction accuracy, with the calibration set achieving an multi-target coefficient of determination (MultiR2) of 0.9992 and multi-target root mean square error (MultiRMSE) of 0.0240, while the prediction set yielded an MultiR2 of 0.9797 and MultiRMSE of 0.1181. Prediction set R2 values for fat, protein, and lactose were 0.9658, 0.9910, and 0.9825, respectively. The proposed method demonstrates robust predictive accuracy and reliability in milk quality assessment, and its potential for application in broader food quality assessments is substantial. PRACTICAL APPLICATION: This study provides a rapid, non-destructive method for assessing milk quality by detecting key nutritional components through hyperspectral imaging, combined with MA-FD preprocessing, ICOA for band selection, and CatBoost for multi-target regression. This approach offers the dairy industry a reliable, non-invasive solution that supports quality control and helps safeguard consumer health.

Rapid Non-Destructive Detection of Rice Seed Vigor via Terahertz Spectroscopy

Rice seed vigor significantly impacts yield, making the selection of high-vigor seeds crucial for agricultural production. Traditional methods for assessing seed vigor are time-consuming and destructive. This study aimed to develop a rapid, non-destructive method for evaluating rice seed vigor using terahertz spectroscopy. Rice seeds with varying vigor levels were prepared through high-temperature and high-humidity aging and classified into high-, low-, and non-vigorous groups based on germination performance. Terahertz transmission imaging (0.1–3 THz) was conducted on 420 seeds, and spectral data were preprocessed using several advanced data processing techniques, including competitive adaptive reweighting (CARS), uninformative variable elimination (UVE), and principal component analysis (PCA). Three chemometric models, namely random forest (RF), K-nearest neighbors (KNN), and partial least squares–discriminant analysis (PLS-DA), were established. The model based on CARS-KNN after band selection achieved the highest prediction accuracy of 97.14%. The results indicate that terahertz spectroscopy combined with band selection methods provides a reliable, non-destructive approach for rice seed vigor assessment, offering significant potential for agricultural quality control.

A fluorescence approach for assessing the composition of evaporating droplets

Accurate measurement of chemical composition is vital for understanding heat and mass transfer mechanisms in two-phase flows like sprays. This Letter presents an optical method based on laser-induced fluorescence and fluorescence lifetime measurement. While solvatochromism—shifts in the emission and absorption spectra based on solvent type that alter fluorescence intensity—is well-documented for many fluorescent dyes, fluorescence lifetime offers distinct advantages for two-phase flow applications. Unlike intensity-based methods, lifetime measurements exhibit greater resilience to light attenuation and scattering at liquid/gas interfaces and allow for more flexible detection band selection. The technique leverages the fluorescence lifetime of eosin Y, which is sensitive to solvent polarity while remaining unaffected by temperature—a crucial feature for systems with fluctuating temperature and composition. We validate this method using acoustically levitated droplets, showing that eosin Y's fluorescence lifetime can reliably distinguish components in binary mixtures like ethanol/water and isopropanol/water. The high volatility of ethanol and isopropanol causes significant cooling during droplet vaporization, leading to water condensation from the surrounding air. This approach enables detailed analysis of complex evaporation and condensation dynamics.

HybridGT: An Integration of Graph Transformer and LSTM for Effective Hyperspectral Band Selection

ABSTRACT Hyperspectral imagery has a high-dimensional curse due to numerous spectral bands. Band selection (BS) is crucial for efficiently reducing dimensionality, retaining only essential bands containing valuable information. However, deep learning-based techniques have gained more attention through trained networks for band selection. Recently, graph-based learning has been extensively used in hyperspectral imagery, revealing intrinsic data relationships. This article presents a novel hybrid approach for hyperspectral band selection, addressing the curse of dimensionality in hyperspectral imagery (HSI). Integrating Long Short Term Memory (LSTM) and Graph Transformer (GT), the method employs Bi-dimensional Empirical Mode Decomposition (BEMD) for spatial data enhancement. Using transfer learning, we explore a ResNet-50 deep network to identify optimal intrinsic mode functions (IMFs). The final band subset will be obtained by concatenating the features extracted from the graph transformer and LSTM networks from selected IMFs and residual IMF, respectively. The proposed HybridGT-BS technique surpasses state-of-the-art methods in classification accuracy across three well-known HSI datasets – IP-Indian Pines, SA-Salinas, and PU-PaviaU. With the support of experimental results, the proposed technique significantly outperforms the classification accuracy with the best bands of the HSIs.

BSDR: A Data-Efficient Deep Learning-Based Hyperspectral Band Selection Algorithm Using Discrete Relaxation.

Hyperspectral band selection algorithms are crucial for processing high-dimensional data, which enables dimensionality reduction, improves data analysis, and enhances computational efficiency. Among these, attention-based algorithms have gained prominence by ranking bands based on their discriminative capability. However, they require a large number of model parameters, which increases the need for extensive training data. To address this challenge, we propose Band Selection through Discrete Relaxation (BSDR), a novel deep learning-based algorithm. BSDR reduces the number of learnable parameters by focusing solely on the target bands, which are typically far fewer than the original bands, thus resulting in a data-efficient configuration that minimizes training data requirements and reduces training time. The algorithm employs discrete relaxation, transforming the discrete problem of band selection into a continuous optimization task, which enables gradient-based search across the spectral dimension. Through extensive evaluations on three benchmark datasets with varying spectral dimensions and characteristics, BSDR demonstrates superior performance for both regression and classification tasks, achieving up to 25% and 34.6% improvements in overall accuracy, compared to the latest attention-based and traditional algorithms, respectively, while reducing execution time by 96.8% and 97.18%. These findings highlight BSDR's effectiveness in addressing key challenges in hyperspectral band selection.

HSF-IBI: A Universal Framework for Extracting Inter-Beat Interval from Heterogeneous Unobtrusive Sensors.

Heartbeat inter-beat interval (IBI) extraction is a crucial technology for unobtrusive vital sign monitoring, yet its precision and robustness remain challenging. A promising approach is fusing heartbeat signals from different types of unobtrusive sensors. This paper introduces HSF-IBI, a novel and universal framework for unobtrusive IBI extraction using heterogeneous sensor fusion. Specifically, harmonic summation (HarSum) is employed for calculating the average heart rate, which in turn guides the selection of the optimal band selection (OBS), the basic sequential algorithmic scheme (BSAS)-based template group extraction, and the template matching (TM) procedure. The optimal IBIs are determined by evaluating the signal quality index (SQI) for each heartbeat. The algorithm is morphology-independent and can be adapted to different sensors. The proposed algorithm framework is evaluated on a self-collected dataset including 19 healthy participants and an open-source dataset including 34 healthy participants, both containing heterogeneous sensors. The experimental results demonstrate that (1) the proposed framework successfully integrates data from heterogeneous sensors, leading to detection rate enhancements of 6.25 % and 5.21 % on two datasets, and (2) the proposed framework achieves superior accuracy over existing IBI extraction methods, with mean absolute errors (MAEs) of 5.25 ms and 4.56 ms on two datasets.

A Design Strategy for Low-Cost Single/Dual-Band Photodetector: Bulk Heterojunction and Interface Engineering.

Photodetectors (PDs) with band selection functions are highly desirable for the future complex environment. Currently, band-selective PDs are typically realized by integrating various optical filters with broadband PDs or constructing heterojunction with multi-photo-absorbing layers. However, these methods cause increased manufacturing costs and device integration complexity. Here, a novel solution-processed perovskite bulk heterojunction (BHJ) layer (MAPbI3:MA3Bi2I9) is developed as photo-absorbing layer, and an effective interface engineering is proposed to selectively shield photo-generated carriers in the BHJ. By regulating the bias voltage, the tunneling probability of photo-generated carrier of MAPbI3 in the BHJ layer can be easily controlled, enabling band-selective capability. The PD exhibits a superior photo-response in the visible and near-infrared region at the -0.3V, with responsivity of 26.5 and 70.2mAW-1 for 500 and 740nm, respectively. But, at the 0.1V, the PD responds only to the visible region with a photo-response of 3.7mAW-1 (500nm) and maintains a high rejection ratio (R500 nm/R740 nm) of 28.5. This PD also exhibits a fast response speed (rise time: 220µs, fall time: 240µs). This work provides a novel strategy to fabricate a cost-effective multifunctional photodetector for future advanced optoelectronic system.

MultiGO: An unsupervised approach based on multi-objective growth optimizer for hyperspectral image band selection

MultiGO: An unsupervised approach based on multi-objective growth optimizer for hyperspectral image band selection