New brown dwarf candidates in the Pleiades

We have performed deep, wide‐field imaging on a ∼0.4 deg² field in the Pleiades (Melotte 22). The selected field was not yet target of a deep search for low mass stars and brown dwarfs. Our limiting magnitudes are R ∼ 22 mag and I ∼ 20 mag, sufficient to detect brown dwarf candidates down to 40 M_J. We found 197 objects, whose location in the (I, R – I) color magnitude diagram is consistent with the age and the distance of the Pleiades. Using CTK R and I as well as JHK photometry from our data and the 2MASS survey we were able to identify 7 new brown dwarf candidates. We present our data reduction technique, which enables us to resample, calibrate, and co‐add many images by just two steps. We estimate the interstellar extinction and the spectral type from our optical and the NIR data using a two‐dimensional χ² fitting (© 2009 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)     

Cosmic‐ray rejection for single spectroscopic CCD images by means of template matching

This article describes a novel algorithm for cosmic‐ray rejection in single spectroscopic CCD images. This algorithm is based on a variation of template matching. It focuses on identifying those pixels belong to spectra, while other conventional algorithms tried to locate the cosmic‐ray hits directly. The main principle is applying template matching to find suspicious blocks, which is followed by surface patching to locate the legitimate pixels accurately. Therefore, the rest pixels are the ones corrupted by cosmic‐ray hits. Meanwhile, almost all the parameters are automatically extracted from the images. Examples of its performance are given for both simulated and observed images. It shows an advantage of significantly low false alarm rate with relatively high detection rate (© 2009 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)     

A novel multifrequency technique for the detection of point sources in cosmic microwave background maps

In this work we address the problem of simultaneous multifrequency detection of extragalactic point sources in the maps of the cosmic microwave background. We apply a new linear filtering technique, the 'matched matrix filters', that incorporates full spatial information, including the cross-correlation among channels, without making any a priori assumption about the spectral behaviour of the sources. A substantial reduction of the background is achieved thanks to the optimal combination of filtered maps. We describe the new technique in detail and apply it to the detection of radio sources and estimation of their parameters in realistic all-sky Planck simulations at 30, 44, 70 and 100 GHz. Then, we compare the results with the single-frequency approach based on the standard matched filter, in terms of reliability, completeness and flux accuracy of the resulting point source catalogues. The new filters outperform the standard matched filters for all these indexes at 30, 44 and 70 GHz, whereas at 100 GHz both kinds of filters have a similar performance. We find a notable increment of the number of true detections for a fixed reliability level. In particular, for a 95 per cent reliability we practically double the number of detections at 30, 44 and 70 GHz.     

Wavelet domain Bayesian denoising of string signal in the cosmic microwave background

An algorithm is proposed for denoising the signal induced by cosmic strings in the cosmic microwave background. A Bayesian approach is taken, based on modelling the string signal in the wavelet domain with generalized Gaussian distributions. Good performance of the algorithm is demonstrated by simulated experiments at arcminute resolution under noise conditions including primary and secondary cosmic microwave background anisotropies, as well as instrumental noise.     

Automatic morphological classification of galaxy images

We describe an image analysis supervised learning algorithm that can automatically classify galaxy images. The algorithm is first trained using manually classified images of elliptical, spiral and edge-on galaxies. A large set of image features is extracted from each image, and the most informative features are selected using Fisher scores. Test images can then be classified using a simple Weighted Nearest Neighbour rule such that the Fisher scores are used as the feature weights. Experimental results show that galaxy images from Galaxy Zoo can be classified automatically to spiral, elliptical and edge-on galaxies with an accuracy of ∼90 per cent compared to classifications carried out by the author. Full compilable source code of the algorithm is available for free download, and its general-purpose nature makes it suitable for other uses that involve automatic image analysis of celestial objects.     

Adaptive Optics Image Restoration Based on Frame Selection and Multi-frame Blind Deconvolution

Restricted by the observational condition and the hardware, adaptive optics can only make a partial correction of the optical images blurred by atmospheric turbulence. A postprocessing method based on frame selection and multi-frame blind deconvolution is proposed for the restoration of high-resolution adaptive optics images. By frame selection we mean we first make a selection of the degraded (blurred) images for participation in the iterative blind deconvolution calculation, with no need of any a priori knowledge, and with only a positivity constraint. This method has been applied to the restoration of some stellar images observed by the 61-element adaptive optics system installed on the Yunnan Observatory 1.2m telescope. The experimental results indicate that this method can effectively compensate for the residual errors of the adaptive optics system on the image, and the restored image can reach the diffraction-limited quality.     

基于遥感数字图像的电子教学地图的制作

随着现代化教学手段在地理教学中的应用,传统的教学挂图可利用计算机作为辅助工具,制作成电子教学地图。本文论述了利用遥感数字图像制作电子教学地图的指导思想、必要性和可行性,并探讨了遥感数字影像教学地图的制作方法。     

全局局部细节感知条件随机场的高分辨率遥感影像建筑物提取

条件随机场模型由于其较强的上下文信息建模能力,被广泛应用于建筑物提取任务中。然而,面对高分辨率遥感影像丰富的地物信息,基于条件随机场的提取方法存在建筑物边界模糊的问题。本文提出了一种全局局部细节感知条件随机场框架,该框架提出全局局部一体化D-LinkNet,在有效利用多尺度建筑物信息的同时保留局部结构信息,解决了传统条件随机场一元势能丢失边界信息的问题。同时,该框架融合分割先验以缓解建筑物类内光谱差异较大的影响,利用更大尺度的上下文信息来精确提取建筑物,并引入局部类别标记代价从而保持细节信息以获取清晰的建筑物边界。实验结果表明,该框架在WHU卫星和航空数据集上的精度评价指标均优于其他对比方法,其IoU分别达89.82%和91.72%,对于复杂场景下的建筑物信息能够获得较好的提取效果。     

IEU-Net高分辨率遥感影像房屋建筑物提取

房屋建筑物作为人类活动的主要场所,快速准确地将其从高分辨率遥感影像中提取出来,对促进遥感信息在防灾减灾、城镇管理等方面的应用具有重要意义。本文基于深度学习,提出了高分辨率遥感影像房屋建筑物像素级精确提取方法。首先,针对样本图像边缘像素特征不足现象,以U-Net模型为基础提出IEU-Net模型,设计了全新的忽略边缘交叉熵函数IELoss并将其作为损失函数,另外添加Dropout和BN层在避免过拟合的同时提高模型训练速度和鲁棒性。其次,为解决模型特征丰富度有限的问题,引入形态学建筑物指数MBI,与遥感影像RGB波段一同参与到模型的分类过程。最后,在模型预测时与IELoss相对应采用忽略边缘预测策略从而获得最佳建筑物提取结果。实验对比分析表明：本文方法能有效克服样本边缘像素特征不足问题并抑制道路、建筑物阴影对结果的影响,提升高分辨率遥感影像中房屋建筑物的提取精度。     

联合卷积神经网络与集成学习的遥感影像场景分类

针对人工设计的中、低层特征难以实现复杂场景影像的高精度分类以及卷积神经网络依赖大量训练数据等问题，结合迁移学习与集成学习，提出了一种联合卷积神经网络与集成学习的遥感影像场景分类算法。首先基于迁移学习的思想，利用在自然影像数据集上训练好的多个深层卷积神经网络模型作为特征提取器，提取图像多个高度抽象的语义特征；然后构建由Logistic回归和支持向量机组成的Stacking集成模型，对同一图像的多个特征分别训练Logistic模型，将预测概率结果融合构建概率特征；最后利用支持向量机对概率特征训练和预测，得到场景影像的分类结果。利用UCMerced_LandUse和NWPU-RESISC 45两种不同规模的遥感影像数据集进行试验，即使在只有10%的数据作为训练样本情况下，本文方法能够分别达到90.74%和87.21%的分类精度。