Input Band Research Articles

Exploring Factors Affecting the Performance of Neural Network Algorithm for Detecting Clouds, Snow, and Lakes in Sentinel-2 Images

Detecting clouds, snow, and lakes in remote sensing images is vital due to their propensity to obscure underlying surface information and hinder data extraction. In this study, we utilize Sentinel-2 images to implement a two-stage random forest (RF) algorithm for image labeling and delve into the factors influencing neural network performance across six aspects: model architecture, encoder, learning rate adjustment strategy, loss function, input image size, and different band combinations. Our findings indicate the Feature Pyramid Network (FPN) achieved the highest MIoU of 87.14%. The multi-head self-attention mechanism was less effective compared to convolutional methods for feature extraction with small datasets. Incorporating residual connections into convolutional blocks notably enhanced performance. Additionally, employing false-color images (bands 12-3-2) yielded a 4.86% improvement in MIoU compared to true-color images (bands 4-3-2). Notably, variations in model architecture, encoder structure, and input band combination had a substantial impact on performance, with parameter variations resulting in MIoU differences exceeding 5%. These results provide a reference for high-precision segmentation of clouds, snow, and lakes and offer valuable insights for applying deep learning techniques to the high-precision extraction of information from remote sensing images, thereby advancing research in deep neural networks for semantic segmentation.

Memory Augmentation and Non-Local Spectral Attention for Hyperspectral Denoising

In this paper, a novel hyperspectral denoising method is proposed, aiming at restoring clean images from images disturbed by complex noise. Previous denoising methods have mostly focused on exploring the spatial and spectral correlations of hyperspectral data. The performances of these methods are often limited by the effective information of the neighboring bands of the image patches in the spectral dimension, as the neighboring bands often suffer from similar noise interference. On the contrary, this study designed a cross-band non-local attention module with the aim of finding the optimal similar band for the input band. To avoid being limited to neighboring bands, this study also set up a memory library that can remember the detailed information of each input band during denoising training, fully learning the spectral information of the data. In addition, we use dense connected module to extract multi-scale spatial information from images separately. The proposed network is validated on both synthetic and real data. Compared with other recent hyperspectral denoising methods, the proposed method not only demonstrates good performance but also achieves better generalization.

Active interference stations based on radiophoton technology

Problem statement. A modern active interference station (SAP) is built on the basis of digital radio frequency memory (RFR). The CRCP technology allows storing copies of radio frequency signals in digital form, as well as performing their digital processing in order to generate signals with modulation of various types [1]. This technology allows you to create a SAP with the possibility of generating simulated interference (or digital copies of the target). SAP based on this technology has one significant drawback, the band of instant signal processing is limited by the capabilities of an analog-to-digital converter and, in terms of use in electronic warfare equipment, in most cases, does not satisfy the condition - the processing band is equal to the range of input signals [2]. The elimination of this disadvantage is possible in several ways, by increasing the number of channels in proportion to the input signal band to the instantaneous RFP band, by expanding the signal processing band using processing in Nyquist zones. Both ways are not enough effect

Predicting band gaps of ABN3 perovskites: an account from machine learning and first-principle DFT studies.

The present paper is primarily focused on predicting the band gaps of nitride perovskites from machine learning (ML) models. The ML models have been framed from the feature descriptors and band gap values of 1563 inorganic nitride perovskites having formation energies <-0.026 eV and band gaps ranging from ∼1.0 to 3.1 eV. Four supervised ML models such as multi-layer perceptron (MLP), gradient boosted decision tree (GBDT), support vector regression (SVR) and random forest regression (RFR) have been considered to predict the band gaps of the said systems. The accuracy of each model has been tested from mean absolute error, root-mean-square error and determination coefficient R2 values. The bivariate plots between the predicted and input band gaps of the compounds for both the training and test datasets have also been estimated. Additionally, two ABN3-type nitride perovskites CeBN3 (B = Mo, W) have been selected and their electronic band structures and optoelectronic properties have been studied from density functional theory (DFT) calculations. The band gap values of the said compounds have been estimated from DFT calculations at PBE, HSE06, G0W0@PBE, G0W0@HSE06 level of theories. The present study will be helpful in exploring the ML models in predicting the band gaps of nitride perovskites which in turn may bear potential applications in photovoltaic cells and optical luminescent devices.

Attention-Based Semantic Segmentation Networks for Forest Applications

Deforestation remains one of the key concerning activities around the world due to commodity-driven extraction, agricultural land expansion, and urbanization. The effective and efficient monitoring of national forests using remote sensing technology is important for the early detection and mitigation of deforestation activities. Deep learning techniques have been vastly researched and applied to various remote sensing tasks, whereby fully convolutional neural networks have been commonly studied with various input band combinations for satellite imagery applications, but very little research has focused on deep networks with high-resolution representations, such as HRNet. In this study, an optimal semantic segmentation architecture based on high-resolution feature maps and an attention mechanism is proposed to label each pixel of the satellite imagery input for forest identification. The selected study areas are located in Malaysian rainforests, sampled from 2016, 2018, and 2020, downloaded using Google Earth Pro. Only a two-class problem is considered for this study, which is to classify each pixel either as forest or non-forest. HRNet is chosen as the baseline architecture, in which the hyperparameters are optimized before being embedded with an attention mechanism to help the model to focus on more critical features that are related to the forest. Several variants of the proposed methods are validated on 6120 sliced images, whereby the best performance reaches 85.58% for the mean intersection over union and 92.24% for accuracy. The benchmarking analysis also reveals that the attention-embedded high-resolution architecture outperforms U-Net, SegNet, and FC-DenseNet for both performance metrics. A qualitative analysis between the baseline and attention-based models also shows that fewer false classifications and cleaner prediction outputs can be observed in identifying the forest areas.

Efficient Deep Semantic Segmentation for Land Cover Classification Using Sentinel Imagery

Nowadays, different machine learning approaches, either conventional or more advanced, use input from different remote sensing imagery for land cover classification and associated decision making. However, most approaches rely heavily on time-consuming tasks to gather accurate annotation data. Furthermore, downloading and pre-processing remote sensing imagery used to be a difficult and time-consuming task that discouraged policy makers to create and use new land cover maps. We argue that by combining recent improvements in deep learning with the use of powerful cloud computing platforms for EO data processing, specifically the Google Earth Engine, we can greatly facilitate the task of land cover classification. For this reason, we modify an efficient semantic segmentation approach (U-TAE) for a satellite image time series to use, as input, a single multiband image composite corresponding to a specific time range. Our motivation is threefold: (a) to improve land cover classification performance and at the same time reduce complexity by using, as input, satellite image composites with reduced noise created using temporal median instead of the original noisy (due to clouds, calibration errors, etc.) images, (b) to assess performance when using as input different combinations of satellite data, including Sentinel-2, Sentinel-1, spectral indices, and ALOS elevation data, and (c) to exploit channel attention instead of the temporal attention used in the original approach. We show that our proposed modification on U-TAE (mIoU: 57.25%) outperforms three other popular approaches, namely random forest (mIoU: 39.69%), U-Net (mIoU: 55.73%), and SegFormer (mIoU: 53.5%), while also using fewer training parameters. In addition, the evaluation reveals that proper selection of the input band combination is necessary for improved performance.

The use of texture analysis and band transformation on multispectral imagery to map open-pit mines using machine learning

Indonesia has the potential for abundant natural mining resources. The Indonesian government needs to monitor mining activity to maintain environmental sustainability and the availability of mining materials in Indonesia. This study aims to map open mining areas based on remote sensing data. This mapping is one of the actions to support sustainable development goals for ensuring sustainable management and efficient use of natural resources. This study was conducted in Central Bangka Regency, Bangka Belitung Island Province, Indonesia, using the multitemporal Sentinel-2 year of 2020-2021. Gray Level Co-Occurrence Matrix and Principal Component Analysis were applied to improve the input band capability in mapping the distribution of open-pit mining locations. A pixel-based machine learning algorithm, Random Forest, was applied to classify mining and non-mining. Classification using texture analysis and spectral transformation mapped an open mining area of 30.67 km2. Classification using only image bands resulted from a mining area of 18.38 km2. The assessment showed that texture analysis and spectral transformation provided an accuracy of 1.22 % higher than the classification using a direct image input band. Overall, the accuracy obtained by both methods was 96.93 % and 95.71 %. Further research on validation with high-resolution data is still needed.

Snow Coverage Mapping by Learning from Sentinel-2 Satellite Multispectral Images via Machine Learning Algorithms

Snow coverage mapping plays a vital role not only in studying hydrology and climatology, but also in investigating crop disease overwintering for smart agriculture management. This work investigates snow coverage mapping by learning from Sentinel-2 satellite multispectral images via machine-learning methods. To this end, the largest dataset for snow coverage mapping (to our best knowledge) with three typical classes (snow, cloud and background) is first collected and labeled via the semi-automatic classification plugin in QGIS. Then, both random forest-based conventional machine learning and U-Net-based deep learning are applied to the semantic segmentation challenge in this work. The effects of various input band combinations are also investigated so that the most suitable one can be identified. Experimental results show that (1) both conventional machine-learning and advanced deep-learning methods significantly outperform the existing rule-based Sen2Cor product for snow mapping; (2) U-Net generally outperforms the random forest since both spectral and spatial information is incorporated in U-Net via convolution operations; (3) the best spectral band combination for U-Net is B2, B11, B4 and B9. It is concluded that a U-Net-based deep-learning classifier with four informative spectral bands is suitable for snow coverage mapping.

Developing a remote sensing-based ecological index based on improved biophysical features

More and more human activities have caused varying degrees of interference with the ecosystem. As an effective quantitative means of eco-environmental quality, remote sensing has been widely used. The remote sensing ecological index (RSEI) is the most popular ecological evaluation model at present. However, the model is affected by the direction of the eigenvector, and two completely opposite results will appear in the calculation of the model. In the past, researchers often manually judged whether the two results met the expectations according to a priori knowledge to select the appropriate calculation results. With the development of remote sensing big data, applying this artificial discrimination method to eco-environmental monitoring under the background of big data is difficult. Therefore, we test the evolution of the RSEI model eigenvector in time series through large samples. It is found that the results of the two models are completely opposite in space. For any model, changing the order of input bands may result in RSEI also being the opposite. Through the large sample test in different time periods, the eigenvectors of each ecological factor affect the direction of the corresponding RSEI. Therefore, we propose an improved model to judge the characteristic contribution direction of the first principal component by selecting the wet factor, which is less affected by seasonal changes. The model direction can be automatically modified without the intervention of researchers according to subjective experience. The improved model can adapt different periods of RSEI calculation. At the same time, no matter how the input band order changes, the final result direction is correct. The improved model makes it possible to monitor and calculate the eco-environmental quality of remote sensing big data and provides a solid scientific basis for the development of the model through research on the mechanism of the model.

Identification of clays and Fe oxide minerals rich alteration zones using a Landsat 8 image of Pu Sam Cap area, Lai Chau

The hydrothermal alteration zones are the important sign for mineral exploration and can be identified by remote sensing images completely, but this is limited due to the effect of vegetable. We address this problem by a method called “Directed Principal Component Analysis” (DPCA) that involves calculating principal components on two input band ratio images. One ratio is a geological discriminant, confused by the presence of vegetation; the second ratio is chosen for its suitability as a vegetation index. DPCA applied on Landsat 8 image in Pu Sam Cap area, Lai Châu characteristied by argilic alteration, sericite alteration, etc., with the typical minerals like kaolinite, illite, etc., and pyrite, chalcopyrite, magnetite; specularite, etc., The results have identified Fe - rich zones in Bai Bang and Nam Tra areas; clay minerals are concentrated mainly in Nam Tra area and along the main faults. The results are also compared with previous research data and fieldtrip data that shows similarity and feasibility. This paper indicated limitation of Landsat image such as spatial resolution, spectral resolution, etc., when applied in the tropical area.

A Practical Method for High-Resolution Burned Area Monitoring Using Sentinel-2 and VIIRS

Mapping burned areas using satellite imagery has become a subject of extensive research over the past decades. The availability of high-resolution satellite data allows burned area maps to be produced with great detail. However, their increasing spatial resolution is usually not matched by a similar increase in the temporal domain. Moreover, high-resolution data can be a computational challenge. Existing methods usually require downloading and processing massive volumes of data in order to produce the resulting maps. In this work we propose a method to make this procedure fast and yet accurate by leveraging the use of a coarse resolution burned area product, the computation capabilities of Google Earth Engine to pre-process and download Sentinel-2 10-m resolution data, and a deep learning model trained to map the multispectral satellite data into the burned area maps. For a 1500 ha fire our method can generate a 10-m resolution map in about 5 min, using a computer with an 8-core processor and 8 GB of RAM. An analysis of six important case studies located in Portugal, southern France and Greece shows the detailed computation time for each process and how the resulting maps compare to the input satellite data as well as to independent reference maps produced by Copernicus Emergency Management System. We also analyze the feature importance of each input band to the final burned area map, giving further insight about the differences among these events.

Aerial Visual Perception in Smart Farming: Field Study of Wheat Yellow Rust Monitoring

Agriculture is facing severe challenges from crop stresses, threatening its sustainable development and food security. This article exploits aerial visual perception for yellow rust disease monitoring, which seamlessly integrates state-of-the-art techniques and algorithms, including unmanned aerial vehicle sensing, multispectral imaging, vegetation segmentation, and deep learning U-Net. A field experiment is designed by infecting winter wheat with yellow rust inoculum, on top of which multispectral aerial images are captured by DJI Matrice 100 equipped with RedEdge camera. After image calibration and stitching, multispectral orthomosaic is labeled for system evaluation by inspecting high-resolution RGB images taken by Parrot Anafi Drone. The merits of the developed framework drawing spectral-spatial information concurrently are demonstrated by showing improved performance over purely spectral-based classifier by the classical random forest algorithm. Moreover, various network input band combinations are tested, including three RGB bands and five selected spectral vegetation indices, by sequential forward selection strategy of wrapper algorithm.

Input imagery, classifiers, and cloud computing: Insights from multi-temporal LULC mapping in the Cambodian Mekong Delta

ABSTRACT The increased open-access availability of radar and optical satellite imagery has engendered numerous land use and land cover (LULC) analyses combining these data sources. In parallel, cloud computing platforms have enabled a wider community to perform LULC classifications over long periods and large areas. However, an assessment of how the performance of classifiers available on these cloud platforms can be optimized for the use of multi-imagery data has been lacking for multi-temporal LULC approaches. This study provides such an assessment for the supervised classifiers available on the open-access Google Earth Engine platform: Naïve Bayes (NB), Classification and Regression Trees (CART), Random Forest (RF), Gradient Tree Boosting (GTB), and Support Vector Machines (SVM). A multi-temporal LULC analysis using Sentinel-1 and 2 is implemented for a study area in the Mekong Delta. Classifier performance is compared for different combinations of input imagery, band sets, and training datasets. The results show that GTB and RF yield the highest overall accuracies, at 94% and 93%. Combining optical and radar imagery boosts classification accuracy for CART, RF, GTB, and SVM by 10–15 percentage points. Furthermore, it reduces the impact of limited training dataset quality for RF, GTB, and SVM.

Nanopower multiple-input DTMOS OTA and its applications to high-order filters for biomedical systems

A new solution for a low-voltage multiple-input dynamic threshold operational transconductance amplifier (MIDT OTA) for low-frequency signal processing applications is presented. The differential pairs of the proposed OTA combine, for the first time, the multiple-input MOS technique along with the dynamic threshold voltage MOS to achieve a very simple and power-efficient CMOS structure with increased total transconductance. Further, the proposed MIDT OTA has been used to realize a fifth-order low-pass filter (LP) and a sixth-order band-pass filter (BP) using coupled-biquads. The proposed OTA and the filter applications are supplied with 0.5 V and consume 5 nW and 25 nW, respectively. The input band pass filter noise is 82.76 µV, while the dynamic range is 59.5 dB for the third intermodulation distortion (IMD) of 2%. The proposed circuits were designed in a 0.18 µm TSMC technology, and validated by simulations using Cadence platform.

Flood Inundation Mapping Using Synthetic Aperture Radar Data Single Polarization: A Case Study of Flood in Lake Tempe, South Sulawesi-Indonesia

Remote sensing data can be used to help disaster management and environmental management because this data has advantages in terms of speed, is more efficient, can reach large and remote areas. Aside from that, it has consistency in measurement, can make repeated measurements, and has measurable accuracy. Floods in Lake Tempe occur almost every year due to overflowing of Lake Tempe. This research will detect flood inundation from Sentinel-1 data with single-polarization of flooding in Lake Tempe, South Sulawesi. The data used were Sentinel-1 data before flooding (2 May 2018) and after flooding (26 May 2018). The Single polarization of the Sentinel-1 can be used very well to identify floods event. The difference of input band uses the threshold of -6, and enables smoothing of 10. Utilization of tidal junction between Lake Tempe and Lake Sidenreng for agriculture and settlement has caused flooding due to lake water overflowing in the rainy season. This single polarization method can already be used in flooded areas, although in some locations it is still overestimates.

A Ku-Band RF Front-End Employing Broadband Impedance Matching with 3.5 dB NF and 21 dB Conversion Gain in 45-nm CMOS Technology

This paper presents a K u -band RF receiver front-end with broadband impedance matching and amplification. The major building blocks of the proposed receiver front-end include a wideband low-noise amplifier (LNA) employing a cascade of resistive feedback inverter (RFI) and transformer-loaded common source amplifier, a down-conversion mixer with push–pull transconductor and complementary LO switching stage, and an output buffer. Push–pull architecture is employed extensively to maximize the power efficiency, bandwidth, and linearity. The proposed two-stage LNA employs the stagger-tuned frequency response in order to extend the RF bandwidth coverage. The input impedance of RFI is carefully analyzed, and a wideband input matching circuit incorporating only a single inductor is presented along with useful equivalent impedance matching models and detailed design analysis. The prototype chip was fabricated in 45-nm CMOS technology and dissipates 78 mW from a 1.2-V supply while occupying chip area of 0.29 mm 2 . The proposed receiver front-end provides 21 dB conversion gain with 7 GHz IF bandwidth, 3.5 dB NF, −15.7 dBm IIP 3 while satisfying &lt;−10 dB input matching over the whole input band.

Delta-sigma modulator based compact sensor signal acquisition front-end system

The proposed delta-sigma modulator (ΔΣM) based signal acquisition architecture uses a differential difference amplifier (DDA) customized for dual purpose roles, namely as instrumentation amplifier and as integrator of ΔΣM. The DDA also provides balanced high input impedance for signal from sensors. Further, programmable input amplification is obtained by adjustment of ΔΣM feedback voltage. Implementation of other functionalities, such as filtering and digitization have also been incorporated. At circuit level, a difference of transconductance of DDA input pairs has been proposed to reduce the effect of input resistor thermal noise of front-end R–C integrator of the ΔΣM. Besides, chopping has been used for minimizing effect of flicker noise. The resulting architecture is an aggregation of functions of entire signal acquisition system within the single block of ΔΣM, and is useful for a multitude of dc-to-medium frequency sensing and similar applications that require high precision at reduced size and power. An implementation of this in 0.18-μm CMOS process has been presented, yielding a simulated peak signal-to-noise ratio of 80 ​dB and dynamic range of 109 ​dBFS in an input signal band of 1 ​kHz while consuming 100 ​μW of power; with the measured signal-to-noise ratio being lower by about 9 ​dB.

Low-Noise Sis Receivers for New Radio-Astronomy Projects

We have developed, manufactured, and tested a waveguide mixer in the range 211–275 GHz on the basis of the superconductor–insulator–superconductor (SIS) tunnel structures. The methods of manufacturing high-quality tunnel structures on quartz substrates have been worked out. To extend the receiver band, the Nb/AlOx/Nb and Nb/AlN/NbN tunnel junctions with a high current density of up to 20 kA/cm2 are employed. The dependence of the characteristics of the receiving elements on the signal frequency is simulated for the intermediate-frequency band 4–12 GHz. The measurements demonstrate a good agreement of the input band of the receiving structures with the calculated results. The uncorrected noise temperature of the receiver amounts to 24 K at a frequency of 265 GHz, which is only two times higher than the quantum limit. The receivers under development are intended for a number of newly-built ground-based radio telescopes (“Suffa” and LLAMA), as well as for the “Millimetron” space program.

Medical Image Fusion With Parameter-Adaptive Pulse Coupled Neural Network in Nonsubsampled Shearlet Transform Domain

As an effective way to integrate the information contained in multiple medical images with different modalities, medical image fusion has emerged as a powerful technique in various clinical applications such as disease diagnosis and treatment planning. In this paper, a new multimodal medical image fusion method in nonsubsampled shearlet transform (NSST) domain is proposed. In the proposed method, the NSST decomposition is first performed on the source images to obtain their multiscale and multidirection representations. The high-frequency bands are fused by a parameter-adaptive pulse-coupled neural network (PA-PCNN) model, in which all the PCNN parameters can be adaptively estimated by the input band. The low-frequency bands are merged by a novel strategy that simultaneously addresses two crucial issues in medical image fusion, namely, energy preservation and detail extraction. Finally, the fused image is reconstructed by performing inverse NSST on the fused high-frequency and low-frequency bands. The effectiveness of the proposed method is verified by four different categories of medical image fusion problems [computed tomography (CT) and magnetic resonance (MR), MR-T1 and MR-T2, MR and positron emission tomography, and MR and single-photon emission CT] with more than 80 pairs of source images in total. Experimental results demonstrate that the proposed method can obtain more competitive performance in comparison to nine representative medical image fusion methods, leading to state-of-the-art results on both visual quality and objective assessment.

MULTI-DATE SENTINEL1 SAR IMAGE TEXTURES DISCRIMINATE PERENNIAL AGROFORESTS IN A TROPICAL FOREST-SAVANNAH TRANSITION LANDSCAPE

Abstract. Synthetic Aperture Radar (SAR) provides consistent information on target land features; especially in tropical conditions that restrain penetration of optical imaging sensors. Because radar response signal is influenced by geometric and di-electrical properties of surface features’, the different land cover may appear similar in radar images. For discriminating perennial cocoa agroforestry land cover, we compare a multi-spectral optical image from RapidEye, acquired in the dry season, and multi-seasonal C-band SAR of Sentinel 1: A final set of 10 (out of 50) images that represent six dry and four wet seasons from 2015 to 2017. We ran eight RF models for different input band combinations; multi-spectral reflectance, vegetation indices, co-(VV) and cross-(VH) polarised SAR intensity and Grey Level Co-occurrence Matrix (GLCM) texture measures. Following a pixel-based image analysis, we evaluated accuracy metrics and uncertainty Shannon entropy. The model comprising co- and cross-polarised texture bands had the highest accuracy of 88.07 % (95 % CI: 85.52–90.31) and kappa of 85.37; and the low class uncertainty for perennial agroforests and transition forests. The optical image had low classification uncertainty for the entire image; but, it performed better in discriminating non-vegetated areas. The measured uncertainty provides reliable validation for comparing class discrimination from different image resolution. The GLCM texture measures that are crucial in delineating vegetation cover differed for the season and polarization of SAR image. Given the high accuracies of mapping, our approach has value for landscape monitoring; and, an improved valuation of agroforestry contribution to REDD+ strategies in the Congo basin sub-region.