Astronomical spectroscopy is crucial for exploring the physical properties,chemical composition,and kinematic behavior of celestial objects.With continuous advancements in observational technology,astronomical spectro...Astronomical spectroscopy is crucial for exploring the physical properties,chemical composition,and kinematic behavior of celestial objects.With continuous advancements in observational technology,astronomical spectroscopy faces the dual challenges of rapidly expanding data volumes and relatively lagging data processing capabilities.In this context,the rise of artificial intelligence technologies offers an innovative solution to address these challenges.This paper analyzes the latest developments in the application of machine learning for astronomical spectral data mining and discusses future research directions in AI-based spectral studies.However,the application of machine learning technologies presents several challenges.The high complexity of models often comes with insufficient interpretability,complicating scientific understanding.Moreover,the large-scale computational demands place higher requirements on hardware resources,leading to a significant increase in computational costs.AI-based astronomical spectroscopy research should advance in the following key directions.First,develop efficient data augmentation techniques to enhance model generalization capabilities.Second,explore more interpretable model designs to ensure the reliability and transparency of scientific conclusions.Third,optimize computational efficiency and reduce the threshold for deep-learning applications through collaborative innovations in algorithms and hardware.Furthermore,promoting the integration of cross-band data processing is essential to achieve seamless integration and comprehensive analysis of multi-source data,providing richer,multidimensional information to uncover the mysteries of the universe.展开更多
The physical and chemical heterogeneities of soils make the soil spectral different and complicated, and it is valuable to increase the accuracy of prediction models for soil organic matter（SOM） based on pre-classif...The physical and chemical heterogeneities of soils make the soil spectral different and complicated, and it is valuable to increase the accuracy of prediction models for soil organic matter（SOM） based on pre-classification. This experiment was conducted under a controllable environment, and different soil samples from northeast of China were measured using ASD2500 hyperspectral instrument. The results showed that there are different reflectances in different soil types. There are statistically significant correlation between SOM and reflectence at 0.05 and 0.01 levels in 550–850 nm, and all soil types get significant at 0.01 level in 650–750 nm. The results indicated that soil types of the northeast can be divided into three categories： The first category shows relatively flat and low reflectance in the entire band; the second shows that the spectral reflectance curve raises fastest in 460–610 nm band, the sharp increase in the slope, but uneven slope changes; the third category slowly uplifts in the visible band, and its slope in the visible band is obviously higher than the first category. Except for the classification by curve shapes of reflectance, principal component analysis is one more effective method to classify soil types. The first principal component includes 62.13–97.19% of spectral information and it mainly relates to the information in 560–600, 630–690 and 690–760 nm. The second mainly represents spectral information in 1 640–1 740, 2 050–2 120 and 2 200–2 300 nm. The samples with high OM are often in the left, and the others with low OM are in the right of the scatter plot（the first principal component is the horizontal axis and the second is the longitudinal axis）. Soil types in northeast of China can be classified effectively by those two principles; it is also a valuable reference to other soil in other areas.展开更多
In some image classification tasks, similarities among different categories are different and the samples are usually misclassified as highly similar categories. To distinguish highly similar categories, more specific...In some image classification tasks, similarities among different categories are different and the samples are usually misclassified as highly similar categories. To distinguish highly similar categories, more specific features are required so that the classifier can improve the classification performance. In this paper, we propose a novel two-level hierarchical feature learning framework based on the deep convolutional neural network(CNN), which is simple and effective. First, the deep feature extractors of different levels are trained using the transfer learning method that fine-tunes the pre-trained deep CNN model toward the new target dataset. Second, the general feature extracted from all the categories and the specific feature extracted from highly similar categories are fused into a feature vector. Then the final feature representation is fed into a linear classifier. Finally, experiments using the Caltech-256, Oxford Flower-102, and Tasmania Coral Point Count(CPC) datasets demonstrate that the expression ability of the deep features resulting from two-level hierarchical feature learning is powerful. Our proposed method effectively increases the classification accuracy in comparison with flat multiple classification methods.展开更多
基金supported by the National Key R&D Program of China(2021YFC2203502 and 2022YFF0711502)the National Natural Science Foundation of China(NSFC)(12173077)+4 种基金the Tianshan Talent Project of Xinjiang Uygur Autonomous Region(2022TSYCCX0095 and 2023TSYCCX0112)the Scientific Instrument Developing Project of the Chinese Academy of Sciences(PTYQ2022YZZD01)China National Astronomical Data Center(NADC)the Operation,Maintenance and Upgrading Fund for Astronomical Telescopes and Facility Instruments,budgeted from the Ministry of Finance of China(MOF)and administrated by the Chinese Academy of SciencesNatural Science Foundation of Xinjiang Uygur Autonomous Region(2022D01A360).
文摘Astronomical spectroscopy is crucial for exploring the physical properties,chemical composition,and kinematic behavior of celestial objects.With continuous advancements in observational technology,astronomical spectroscopy faces the dual challenges of rapidly expanding data volumes and relatively lagging data processing capabilities.In this context,the rise of artificial intelligence technologies offers an innovative solution to address these challenges.This paper analyzes the latest developments in the application of machine learning for astronomical spectral data mining and discusses future research directions in AI-based spectral studies.However,the application of machine learning technologies presents several challenges.The high complexity of models often comes with insufficient interpretability,complicating scientific understanding.Moreover,the large-scale computational demands place higher requirements on hardware resources,leading to a significant increase in computational costs.AI-based astronomical spectroscopy research should advance in the following key directions.First,develop efficient data augmentation techniques to enhance model generalization capabilities.Second,explore more interpretable model designs to ensure the reliability and transparency of scientific conclusions.Third,optimize computational efficiency and reduce the threshold for deep-learning applications through collaborative innovations in algorithms and hardware.Furthermore,promoting the integration of cross-band data processing is essential to achieve seamless integration and comprehensive analysis of multi-source data,providing richer,multidimensional information to uncover the mysteries of the universe.
基金supported by the National Natural Science Foundation of China(41371292)
文摘The physical and chemical heterogeneities of soils make the soil spectral different and complicated, and it is valuable to increase the accuracy of prediction models for soil organic matter（SOM） based on pre-classification. This experiment was conducted under a controllable environment, and different soil samples from northeast of China were measured using ASD2500 hyperspectral instrument. The results showed that there are different reflectances in different soil types. There are statistically significant correlation between SOM and reflectence at 0.05 and 0.01 levels in 550–850 nm, and all soil types get significant at 0.01 level in 650–750 nm. The results indicated that soil types of the northeast can be divided into three categories： The first category shows relatively flat and low reflectance in the entire band; the second shows that the spectral reflectance curve raises fastest in 460–610 nm band, the sharp increase in the slope, but uneven slope changes; the third category slowly uplifts in the visible band, and its slope in the visible band is obviously higher than the first category. Except for the classification by curve shapes of reflectance, principal component analysis is one more effective method to classify soil types. The first principal component includes 62.13–97.19% of spectral information and it mainly relates to the information in 560–600, 630–690 and 690–760 nm. The second mainly represents spectral information in 1 640–1 740, 2 050–2 120 and 2 200–2 300 nm. The samples with high OM are often in the left, and the others with low OM are in the right of the scatter plot（the first principal component is the horizontal axis and the second is the longitudinal axis）. Soil types in northeast of China can be classified effectively by those two principles; it is also a valuable reference to other soil in other areas.
基金Project supported by the National Natural Science Foundation of China(No.61379074)the Zhejiang Provincial Natural Science Foundation of China(Nos.LZ12F02003 and LY15F020035)
文摘In some image classification tasks, similarities among different categories are different and the samples are usually misclassified as highly similar categories. To distinguish highly similar categories, more specific features are required so that the classifier can improve the classification performance. In this paper, we propose a novel two-level hierarchical feature learning framework based on the deep convolutional neural network(CNN), which is simple and effective. First, the deep feature extractors of different levels are trained using the transfer learning method that fine-tunes the pre-trained deep CNN model toward the new target dataset. Second, the general feature extracted from all the categories and the specific feature extracted from highly similar categories are fused into a feature vector. Then the final feature representation is fed into a linear classifier. Finally, experiments using the Caltech-256, Oxford Flower-102, and Tasmania Coral Point Count(CPC) datasets demonstrate that the expression ability of the deep features resulting from two-level hierarchical feature learning is powerful. Our proposed method effectively increases the classification accuracy in comparison with flat multiple classification methods.
