Coded aperture snapshot spectral imaging(CASSI) has been discussed in recent years. It has the remarkable advantages of high optical throughput, snapshot imaging, etc. The entire spatial-spectral data-cube can be reco...Coded aperture snapshot spectral imaging(CASSI) has been discussed in recent years. It has the remarkable advantages of high optical throughput, snapshot imaging, etc. The entire spatial-spectral data-cube can be reconstructed with just a single two-dimensional(2D) compressive sensing measurement. On the other hand, for less spectrally sparse scenes,the insufficiency of sparse sampling and aliasing in spatial-spectral images reduce the accuracy of reconstructed threedimensional(3D) spectral cube. To solve this problem, this paper extends the improved CASSI. A band-pass filter array is mounted on the coded mask, and then the first image plane is divided into some continuous spectral sub-band areas. The entire 3D spectral cube could be captured by the relative movement between the object and the instrument. The principle analysis and imaging simulation are presented. Compared with peak signal-to-noise ratio(PSNR) and the information entropy of the reconstructed images at different numbers of spectral sub-band areas, the reconstructed 3D spectral cube reveals an observable improvement in the reconstruction fidelity, with an increase in the number of the sub-bands and a simultaneous decrease in the number of spectral channels of each sub-band.展开更多
Most unsupervised or semisupervised hyperspectral anomaly detection(HAD)methods train background reconstruction models in the original spectral domain.However,due to the noise and spatial resolution limitations,there ...Most unsupervised or semisupervised hyperspectral anomaly detection(HAD)methods train background reconstruction models in the original spectral domain.However,due to the noise and spatial resolution limitations,there may be a lack of discrimination between backgrounds and anomalies.This makes it easy for the autoencoder to capture the lowlevel features shared between the two,thereby increasing the difficulty of separating anomalies from the backgrounds,which runs counter to the purpose of HAD.To this end,the authors map the original spectrums to the fractional Fourier domain(FrFD)and reformulate it as a mapping task in which restoration errors are employed to distinguish background and anomaly.This study proposes a novel frequency‐to‐spectrum mapping generative adversarial network for HAD.Specifically,the depth separable features of backgrounds and anomalies are enhanced in the FrFD.Due to the semisupervised approach,FTSGAN needs to learn the embedded features of the backgrounds,thus mapping and restoring them from the FrFD to the original spectral domain.This strategy effectively prevents the model from focussing on the numerical equivalence of input and output,and restricts the ability of FTSGAN to restore anomalies.The comparison and analysis of the experiments verify that the proposed method is competitive.展开更多
Recently,deep learning has achieved considerable results in the hyperspectral image(HSI)classification.However,most available deep networks require ample and authentic samples to better train the models,which is expen...Recently,deep learning has achieved considerable results in the hyperspectral image(HSI)classification.However,most available deep networks require ample and authentic samples to better train the models,which is expensive and inefficient in practical tasks.Existing few‐shot learning(FSL)methods generally ignore the potential relationships between non‐local spatial samples that would better represent the underlying features of HSI.To solve the above issues,a novel deep transformer and few‐shot learning(DTFSL)classification framework is proposed,attempting to realize fine‐grained classification of HSI with only a few‐shot instances.Specifically,the spatial attention and spectral query modules are introduced to overcome the constraint of the convolution kernel and consider the information between long‐distance location(non‐local)samples to reduce the uncertainty of classes.Next,the network is trained with episodes and task‐based learning strategies to learn a metric space,which can continuously enhance its modelling capability.Furthermore,the developed approach combines the advantages of domain adaptation to reduce the variation in inter‐domain distribution and realize distribution alignment.On three publicly available HSI data,extensive experiments have indicated that the proposed DT‐FSL yields better results concerning state‐of‐the‐art algorithms.展开更多
Arbitrary‐oriented object detection is widely used in aerial image applications because of its efficient object representation.However,the use of oriented bounding box aggravates the imbalance between positive and ne...Arbitrary‐oriented object detection is widely used in aerial image applications because of its efficient object representation.However,the use of oriented bounding box aggravates the imbalance between positive and negative samples when using one‐stage object detectors,which seriously decreases the detection accuracy.We believe that it is the anchor learning strategy(ALS)used by such detectors that needs to take the responsibility.In this study,three perspectives on ALS design were summarised and ALS—Performance Releaser with Smart Anchor Learning(PRSAL)was proposed.Performance Releaser with Smart Anchor Learning is a dynamic ALS that utilises anchor classification ability as an equivalent indicator to anchor box regression ability,this allows anchors with high detection potential to be filtered out in a more reasonable way.At the same time,PRSAL focuses more on anchor potential and it is able to automatically select a number of positive samples that far exceed that of other methods by activating anchors that previously had a low spatial overlap,thereby releasing the detection performance.We validate the PRSAL using three remote sensing datasets—HRSC2016,DOTA and UCAS‐AOD as well as one scene text dataset—ICDAR 2013.The experimental results show that the proposed method gives substantially better results than existing models.展开更多
The conventional optical system design employs combinations of different lenses to combat aberrations, which usually leads to considerable volume and weight. In this Letter, a tailored design scheme that exploits stat...The conventional optical system design employs combinations of different lenses to combat aberrations, which usually leads to considerable volume and weight. In this Letter, a tailored design scheme that exploits state-of-the-art digital aberration correction algorithms in addition to traditional optics design is investigated. In particular, the proposed method is applied to the design of refractive telescopes by shifting the burden of correcting chromatic aberrations to software. By enforcing cross-channel information transfer in a post-processing step, the uncorrected chromatic aberrations are well-mitigated. Accordingly, a telescope of F-8, 1400 mm focal length, and 0.14° field of view is designed with only two lens elements. The image quality of the designed telescope is evaluated by comparing it to the equivalent designs with multiple lenses in a traditional optical design manner, which validates the effectiveness of our design scheme.展开更多
Injection moulding has shown its advantages and prevalence in the production of plastic optical components,the performance and functionality of which rely on the precision replication of surface forms and on minimizin...Injection moulding has shown its advantages and prevalence in the production of plastic optical components,the performance and functionality of which rely on the precision replication of surface forms and on minimizing residual stress.The present work constitutes a systematic and comprehensive analysis of practical microlens arrays that are designed for light-field applications.Process parameters are screened and optimized using a two-stage design of experiments approach.Based on in-line process monitoring and a quantitative and qualitative evaluation being carried out in terms of geometric accuracy,surface quality and stress birefringence,the replication is shown to relate directly to machine settings and dynamic machine responses.The geometric accuracy and stress birefringence are both largely associated with screw displacement and peak cavity pressure during the packing stage,while surface quality is closely related to cavity temperature.This study provides important insights and recommendations regarding the overall replication quality of microlens arrays,while advanced injection moulding solutions may be necessary to further improve the general replication quality.展开更多
Hyperspectral remote sensing image(HSI)fusion with multispectral remote sensing images(MSI)improves data resolution.However,current fusion algorithms focus on local information and overlook long-range dependencies.The...Hyperspectral remote sensing image(HSI)fusion with multispectral remote sensing images(MSI)improves data resolution.However,current fusion algorithms focus on local information and overlook long-range dependencies.The parameter of network tuning prioritizes global optimization,neglecting spatial and spectral constraints,and limiting spatial and spectral reconstruction capabilities.This study introduces SwinGAN,a fusion network combining Swin Transformer,CNN,and GAN architectures.SwinGAN’s generator employs a detail injection framework to separately extract HSI and MSI features,fusing them to generate spatial residuals.These residuals are injected into the supersampled HSI to produce thefinal image,while a pure CNN architecture acts as the discriminator,enhancing the fusion quality.Additionally,we introduce a new adaptive loss function that improves image fusion accuracy.The loss function uses L1 loss as the content loss,and spatial and spectral gradient loss functions are introduced to improve the spatial representation and spectralfidelity of the fused images.Our experimental results on several datasets demonstrate that SwinGAN outperforms current popular algorithms in both spatial and spectral reconstruction capabilities.The ablation experiments also demonstrate the rationality of the various components of the proposed loss function.展开更多
Despite tons of advanced classification models that have recently been developed for the land cover mapping task,the monotonicity of a single remote sensing data source,such as only using hyperspectral data or multisp...Despite tons of advanced classification models that have recently been developed for the land cover mapping task,the monotonicity of a single remote sensing data source,such as only using hyperspectral data or multispectral data,hinders the classification accuracy from being further improved and tends to meet the performance bottleneck.For this reason,we develop a novel superpixel-based subspace learning model,called Supace,by jointly learning multimodal feature representations from HS and MS superpixels for more accurate LCC results.Supace can learn a common subspace across multimodal RS data,where the diverse and complementary information from different modalities can be better combined,being capable of enhancing the discriminative ability of to-be-learned features in a more effective way.To better capture semantic information of objects in the feature learning process,superpixels that beyond pixels are regarded as the study object in our Supace for LCC.Extensive experiments have been conducted on two popular hyperspectral and multispectral datasets,demonstrating the superiority of the proposed Supace in the land cover classification task compared with several well-known baselines related to multimodal remote sensing image feature learning.展开更多
基金supported by the National Natural Science Foundation for Distinguished Young Scholars of China(Grant No.61225024)the National High Technology Research and Development Program of China(Grant No.2011AA7012022)
文摘Coded aperture snapshot spectral imaging(CASSI) has been discussed in recent years. It has the remarkable advantages of high optical throughput, snapshot imaging, etc. The entire spatial-spectral data-cube can be reconstructed with just a single two-dimensional(2D) compressive sensing measurement. On the other hand, for less spectrally sparse scenes,the insufficiency of sparse sampling and aliasing in spatial-spectral images reduce the accuracy of reconstructed threedimensional(3D) spectral cube. To solve this problem, this paper extends the improved CASSI. A band-pass filter array is mounted on the coded mask, and then the first image plane is divided into some continuous spectral sub-band areas. The entire 3D spectral cube could be captured by the relative movement between the object and the instrument. The principle analysis and imaging simulation are presented. Compared with peak signal-to-noise ratio(PSNR) and the information entropy of the reconstructed images at different numbers of spectral sub-band areas, the reconstructed 3D spectral cube reveals an observable improvement in the reconstruction fidelity, with an increase in the number of the sub-bands and a simultaneous decrease in the number of spectral channels of each sub-band.
基金supported by the National Natural Science Foundation of China under Grant 62161160336Grant 41871245in part by the Belgium Vlaio project(AI ICON‐2021‐0599:Smart industrial spectral cameras via artificial intelligence).
文摘Most unsupervised or semisupervised hyperspectral anomaly detection(HAD)methods train background reconstruction models in the original spectral domain.However,due to the noise and spatial resolution limitations,there may be a lack of discrimination between backgrounds and anomalies.This makes it easy for the autoencoder to capture the lowlevel features shared between the two,thereby increasing the difficulty of separating anomalies from the backgrounds,which runs counter to the purpose of HAD.To this end,the authors map the original spectrums to the fractional Fourier domain(FrFD)and reformulate it as a mapping task in which restoration errors are employed to distinguish background and anomaly.This study proposes a novel frequency‐to‐spectrum mapping generative adversarial network for HAD.Specifically,the depth separable features of backgrounds and anomalies are enhanced in the FrFD.Due to the semisupervised approach,FTSGAN needs to learn the embedded features of the backgrounds,thus mapping and restoring them from the FrFD to the original spectral domain.This strategy effectively prevents the model from focussing on the numerical equivalence of input and output,and restricts the ability of FTSGAN to restore anomalies.The comparison and analysis of the experiments verify that the proposed method is competitive.
基金supported by the National Natural Science Foundation of China under Grant 62161160336 and Grant 42030111.
文摘Recently,deep learning has achieved considerable results in the hyperspectral image(HSI)classification.However,most available deep networks require ample and authentic samples to better train the models,which is expensive and inefficient in practical tasks.Existing few‐shot learning(FSL)methods generally ignore the potential relationships between non‐local spatial samples that would better represent the underlying features of HSI.To solve the above issues,a novel deep transformer and few‐shot learning(DTFSL)classification framework is proposed,attempting to realize fine‐grained classification of HSI with only a few‐shot instances.Specifically,the spatial attention and spectral query modules are introduced to overcome the constraint of the convolution kernel and consider the information between long‐distance location(non‐local)samples to reduce the uncertainty of classes.Next,the network is trained with episodes and task‐based learning strategies to learn a metric space,which can continuously enhance its modelling capability.Furthermore,the developed approach combines the advantages of domain adaptation to reduce the variation in inter‐domain distribution and realize distribution alignment.On three publicly available HSI data,extensive experiments have indicated that the proposed DT‐FSL yields better results concerning state‐of‐the‐art algorithms.
基金supported by the National Key R&D Program of China(Grant No.2021YFB3900502)the Scientific Research and Development Program of China Railway(K2019G008)the Tianjin Intelligent Manufacturing Special Fund Project(No.20201198).
文摘Arbitrary‐oriented object detection is widely used in aerial image applications because of its efficient object representation.However,the use of oriented bounding box aggravates the imbalance between positive and negative samples when using one‐stage object detectors,which seriously decreases the detection accuracy.We believe that it is the anchor learning strategy(ALS)used by such detectors that needs to take the responsibility.In this study,three perspectives on ALS design were summarised and ALS—Performance Releaser with Smart Anchor Learning(PRSAL)was proposed.Performance Releaser with Smart Anchor Learning is a dynamic ALS that utilises anchor classification ability as an equivalent indicator to anchor box regression ability,this allows anchors with high detection potential to be filtered out in a more reasonable way.At the same time,PRSAL focuses more on anchor potential and it is able to automatically select a number of positive samples that far exceed that of other methods by activating anchors that previously had a low spatial overlap,thereby releasing the detection performance.We validate the PRSAL using three remote sensing datasets—HRSC2016,DOTA and UCAS‐AOD as well as one scene text dataset—ICDAR 2013.The experimental results show that the proposed method gives substantially better results than existing models.
基金supported by the Joint Foundation Program of the Chinese Academy of Sciences for Equipment Pre-Feasibility Study(No.6141A01011601)the National Natural Science Foundation of China(Nos.61775219 and61640422)the Equipment Research Program of the Chinese Academy of Sciences(No.Y70X25A1HY)
文摘The conventional optical system design employs combinations of different lenses to combat aberrations, which usually leads to considerable volume and weight. In this Letter, a tailored design scheme that exploits state-of-the-art digital aberration correction algorithms in addition to traditional optics design is investigated. In particular, the proposed method is applied to the design of refractive telescopes by shifting the burden of correcting chromatic aberrations to software. By enforcing cross-channel information transfer in a post-processing step, the uncorrected chromatic aberrations are well-mitigated. Accordingly, a telescope of F-8, 1400 mm focal length, and 0.14° field of view is designed with only two lens elements. The image quality of the designed telescope is evaluated by comparing it to the equivalent designs with multiple lenses in a traditional optical design manner, which validates the effectiveness of our design scheme.
基金The support from the National Key Research&Development Program(Grant No.2016YFB1102203)the China Scholarship Council,the National Natural Science Foundation of China(Grant No.61675149)Science Foundation Ireland(Grant No.15/RP/B3208)is gratefully acknowledged.
文摘Injection moulding has shown its advantages and prevalence in the production of plastic optical components,the performance and functionality of which rely on the precision replication of surface forms and on minimizing residual stress.The present work constitutes a systematic and comprehensive analysis of practical microlens arrays that are designed for light-field applications.Process parameters are screened and optimized using a two-stage design of experiments approach.Based on in-line process monitoring and a quantitative and qualitative evaluation being carried out in terms of geometric accuracy,surface quality and stress birefringence,the replication is shown to relate directly to machine settings and dynamic machine responses.The geometric accuracy and stress birefringence are both largely associated with screw displacement and peak cavity pressure during the packing stage,while surface quality is closely related to cavity temperature.This study provides important insights and recommendations regarding the overall replication quality of microlens arrays,while advanced injection moulding solutions may be necessary to further improve the general replication quality.
基金supported by the National Key Research and Development Program of China(No.2020YFA0714103).
文摘Hyperspectral remote sensing image(HSI)fusion with multispectral remote sensing images(MSI)improves data resolution.However,current fusion algorithms focus on local information and overlook long-range dependencies.The parameter of network tuning prioritizes global optimization,neglecting spatial and spectral constraints,and limiting spatial and spectral reconstruction capabilities.This study introduces SwinGAN,a fusion network combining Swin Transformer,CNN,and GAN architectures.SwinGAN’s generator employs a detail injection framework to separately extract HSI and MSI features,fusing them to generate spatial residuals.These residuals are injected into the supersampled HSI to produce thefinal image,while a pure CNN architecture acts as the discriminator,enhancing the fusion quality.Additionally,we introduce a new adaptive loss function that improves image fusion accuracy.The loss function uses L1 loss as the content loss,and spatial and spectral gradient loss functions are introduced to improve the spatial representation and spectralfidelity of the fused images.Our experimental results on several datasets demonstrate that SwinGAN outperforms current popular algorithms in both spatial and spectral reconstruction capabilities.The ablation experiments also demonstrate the rationality of the various components of the proposed loss function.
基金supported by the National Natural Science Foundation of China (Grant Nos. 62161160336, 42030111, and 62101045)the China Postdoctoral Science Foundation Funded Project (Grant No. 2021M690385)
文摘Despite tons of advanced classification models that have recently been developed for the land cover mapping task,the monotonicity of a single remote sensing data source,such as only using hyperspectral data or multispectral data,hinders the classification accuracy from being further improved and tends to meet the performance bottleneck.For this reason,we develop a novel superpixel-based subspace learning model,called Supace,by jointly learning multimodal feature representations from HS and MS superpixels for more accurate LCC results.Supace can learn a common subspace across multimodal RS data,where the diverse and complementary information from different modalities can be better combined,being capable of enhancing the discriminative ability of to-be-learned features in a more effective way.To better capture semantic information of objects in the feature learning process,superpixels that beyond pixels are regarded as the study object in our Supace for LCC.Extensive experiments have been conducted on two popular hyperspectral and multispectral datasets,demonstrating the superiority of the proposed Supace in the land cover classification task compared with several well-known baselines related to multimodal remote sensing image feature learning.
基金supported by the National Key Research and Development Program of China(2020YFA0607900,2020YFA0608003,and 2021YFC3101601)the National Natural Science Foundation of China(42125503 and 42075137)the National Key Scientific and Technological Infrastructure Project‘‘Earth System Science Numerical Simulator Facility”(Earth Lab)。
