In this paper, we present an approach to learn latent semantic analysis models from loosely annotated images for automatic image annotation and indexing. The given annotation in training images is loose due to: 1. ambiguous correspondences between visual features and annotated keywords; 2. incomplete lists of annotated keywords. The second reason motivates us to enrich the incomplete annotation in a simple way before learning a topic model. In particular, some "imagined" keywords are poured into the incomplete annotation through measuring similarity between keywords in terms of their co-occurrence. Then, both given and imagined annotations are employed to learn probabilistic topic models for automatically annotating new images. We conduct experiments on two image databases (i.e., Corel and ESP) coupled with their loose annotations, and compare the proposed method with state-of-the-art discrete annotation methods. The proposed method improves word-driven probability latent semantic analysis (PLSA-words) up to a comparable performance with the best discrete annotation method, while a merit of PLSA-words is still kept, i.e., a wider semantic range.
Best-routing is one of the effective ways to solve the problem of traffic jam in a technical way. Based on the classic Dijkstra Algorithm, the bidirectional search algorithm is adopted to improve algorithmic efficiency in this paper. And section resistance is also adopted in best-routing model in order to indicate not only the condition of road itself, but also the information like traffic flow, which can make the road information more general and efficient. And cross linked list is adopted to reflect topological information of road net, which ensures the weight keep minimum. The feasibility and efficiency are verified by a study case of local area in Beijing.
Vegetation is a fundamental component of urban environment and its abundance is determinant of urban climate and
urban ground energy fluxes. Based on the radiometric normalization of multitemporal ASTER imageries, the objectives
of this study are: firstly, to estimate the vegetation abundance based on linear spectral mixture model (LSMM), and to
compare it with NDVI and SDVI; secondly, to analyze the spatial distribution patterns of urban vegetation abundance in
different seasons combined with some landscape metrics. The result indicates that both the vegetation abundance estimation based on LSMM and SDVI can reach high accuracy; however, NDVI is not a robust parameter for vegetation abundance estimation because there is significant non-linear effect between NDVI and vegetation abundance. This study reveals that the landscape characteristics of vegetation abundance is most complicated in summer, with spring and autumn less complicated and simplest in winter. This provides valuable information for urban vegetation abundance estimation and its seasonal change monitoring using remote sensing data.
In this article, we present some experiments on coral reef benthic cover mapping with fused IKONOS image. The
objective of our study is to establish an efficient approach for the classification task on hand. Four scenarios are designed
and in each scenario two classification methods (Maximum Likelihood and Decision Tree) are implemented. Ground
truth data is obtained through visual interpretation and manual digitization, against which accuracy of classification map
is calculated. Results indicate that mining spectral information deeply (scenario III and IV) can increase classification
accuracy dramatically. Compared with conventional utilization of spectral data (scenarioI), classification accuracy of ML
and DT respectively increases by 3.94% and 5.15% under scenario IV. However, when spectral and spatial information is
combined together (scenario II), accuracy of ML and DT is respectively reduced by 8.02% and 2.31%. It can be
concluded from our study that when classify benthic cover with high-resolution remote sensing data in pixel-based
pattern, utilization of spatial information should not be excessively emphasized. Fully exploiting spectral information
may bring more benefits. Moreover, DT is more robust and can produce more accurate classification results than ML.
Our results help scientists and managers in applying IKONOS-class data for coral reef mapping applications.
Hyperspectral Remote Sensing (HRS) is one of the most significant recent achievements of Earth Observation Technology. Classification is the most commonly employed processing methodology. In this paper three new hyperspectral RS image classification methods are analyzed. These methods are: Object-oriented FIRS image classification, HRS image classification based on information fusion and HSRS image classification by Back Propagation Neural Network (BPNN). OMIS FIRS image is used as the example data. Object-oriented techniques have
gained popularity for RS image classification in recent years. In such method, image segmentation is used to extract the regions from the pixel information based on homogeneity criteria at first, and spectral parameters like mean vector, texture, NDVI and spatial/shape parameters like aspect ratio, convexity, solidity, roundness and orientation for each region are calculated, finally classification of the image using the region feature vectors and also using suitable
classifiers such as artificial neural network (ANN). It proves that object-oriented methods can improve classification accuracy since they utilize information and features both from the point and the neighborhood, and the processing unit is a polygon (in which all pixels are homogeneous and belong to the class). HRS image classification based on information fusion, divides all bands of the image into different groups initially, and extracts features from every group according to the properties of each group. Three levels of information fusion: data level fusion, feature level fusion and decision level fusion are used to HRS image classification. Artificial Neural Network (ANN) can perform well in RS image
classification. In order to promote the advances of ANN used for HIRS image classification, Back Propagation Neural Network (BPNN), the most commonly used neural network, is used to HRS image classification.
The maximum likelihood classification (MLC) is one of the most popular methods in remote sensing image
classification. Because the maximum likelihood classification is based on spectrum of objects, it cannot correctly
distinguish objects that have same spectrum and cannot reach the accuracy requirement. In this paper, we take an area of
Langfang of Hebei province in China as an example and discuss the method of combining texture of panchromatic image
with spectrum to improve the accuracy of CBERS02 CCD image information extraction. Firstly, analysis of the textures
of the panchromatic image (CCD5) made by using texture analysis of Gray Level Coocurrence Matrices and statistic
index. Then optimal texture window size of angular second moment, contrast, entropy and correlation is obtained
according to variation coefficient of each texture measure for each thematic class. The chosen optimal window size is
that from which the value of variation coefficient starts to stabilize while having the smallest value. The output images
generated by texture analysis are used as additional bands together with other multi-spectral bands(CCD1-4) in
classification. Objects that have same spectrums can be distinguished. Finally, the accuracy measurement is compared
with the classification based on spectrum only .The result indicates that the objects with same spectrum are distinguished
by using texture analysis in image classification, and the spectral /textural combination improves more than spectrum
only in classification accuracy.
Spectrum of healthy green vegetation shows idiographic features of "peak and valley", the spectral curve will vary when
crop's biochemical status changes (e.g. disease harmed). Normalized Difference Vegetation Index (NDVI) is an
important vegetation index and has been proved to be very useful to vegetation change detection, vegetation
classification and some parameters calculation. Based on the differences of spectra information and characteristics
between multi-temporal hyperspectral images, a new adjustable vegetation index, Multi-Temporal NDVI (MT-NDVI), is
provided in this paper. Comparing to the classification of Spectral Angle Mapper (SAM), mapping and analysis using
MT-NDVI data can be well utilized for monitoring and recognizing crop disease from multi-temporal airborne PHI
(Pushbroom Hyperspectral Imager) image data acquired at the same field. The applicable result shows that MT-NDVI is
suitable way to extract crop disease information and estimate disease degrees.
The study site is selected in a Malaysian tropical rainforest area, which consists of a mixture of plain, hilly and mountainous terrain. Digital Elevation Model (DEM) images were generated from nine RADARSAT-1 imageries (F, S and W beam modes) which make up six stereo pair combinations. The DEM accuracies for all the stereo combinations have been validated and compared to each other. The results show that numerous factors affect the final DEM accuracy. In flat areas, the final DEM accuracy is highly correlated to the stereo intersection geometry of the different image combinations. The higher the stereo intersection angle of the same beam mode, the better the accuracy of the final DEM.
Spectral similarity measure plays important roles in hyperspectral Remote Sensing (RS) information processing, and it can be used to content-based hyperspectral RSimage retrieval effectively too. The applications of spectral features to Remote Sensing (RS) image retrieval are discussed by taking hyperspectral RS image as examples oriented to the demands of massive information management. It is proposed that spectral features-based image retrieval includes two modes: retrieval based on point template and facial template. Point template is used usually, for example, a spectral curve, or a pixel vector in hyperspectral RS image. One or more regions (or blocks with area shape) are given as examples in image retrieval based on facial template. The most important issues in image retrieval are spectral features extraction and spectral similarity measure. Spectral vector can be used to retrieval directly, and spectral angle and spectral information divergence (SID) are more effective than Euclidean distance and correlation coefficient in similarity measure and image retrieval. Both point and pure area template can be transformed into spectral vector and used to spectral similarity measure. In addition, the local maximum and minimum in reflection spectral curve, corresponding to reflection peak and absorption valley, can be used to retrieval also. The width, height, symmetry and power of each peak or valley can be used to encode spectral features. By comparison to three approaches for spectral absorption and reflection features matching and similarity measures, it is found that spectral absorption and reflection features are not very effective in hyperspectral RS image retrieval. Finally, a prototype system is designed, and it proves that the hyperspectral RS image retrieval based on spectral similarity measure proposed in this paper is effective and some similarity measure index including spectral angle, SID and encoding measure are suitable for image retrieval in practice.
Noises are inevitable in Hyperspectral Remote Sensing (HRS) image, it is very important to design effective filter to reduce the impacts of noises and enhance image quality and information content. Based on the characteristics of HRS image, three filtering strategies, including image dimension filtering, spectral dimension filtering and three-dimensional filtering, are proposed in this paper. The principle of image dimension filtering is similar to traditional image filtering from spatial and frequency domain. The image of each band is viewed as an independent set and filtering operation is used to it. Some filters, including mean filter, medium filter and frequency filter, are used to reduce noises in every band. The key idea of spectral dimension filtering is to take every pixel as the processing target, and the gray value (or albedo) of the pixel on all bands will form a spectral vector. Filter is used to the spectral vector of every pixel, and mean filter with different scales is tested in this paper. Three-dimension filtering is different from the former two methods by its spatial and spectral dimension processing simultaneously. It views HRS image as a large data cube with row, column and layer (band), so filter is based on data cube. In this paper the 3×3×3 cube is used as filtering template, and that means those neighbors of adjacent bands of a pixel on a given band will be used to filter, so both spatial and spectral information is considered in this new method. Finally, some examples are experimented and quality assessment of sole band, similarity measure to some pixels and other statistical indexes are used to assess the performance, and then related conclusions and suggestions are given.
Vegetation fraction, the ratio of vegetation occupying a unit area, as a significant parameter in the development of climate and ecological models, is indispensable information of many global and regional climate numerical models. It is also an important basic data of describing ecosystem. However, It is also a wasting manpower and financial resources with low-precision work to measure the vegetation fraction by fieldwork, especially in large areas. This study explores the potential of deriving vegetation fraction from normalized difference vegetation index (NDVI) using the TM data. Under the assumption that the pixel of TM image is a mosaic structure, sub-pixel models for vegetation fraction estimation have been introduced firstly. Then the idea of utility of different sub-pixel model for vegetation fraction estimation based on land cover classification is proposed. The model for vegetation fraction estimation has been established under many assumptions, and there is the complex relationship of vegetation index vegetation fraction and leaf area index, so it is unrealistic to obtain vegetation fraction with high precision. But it is helpful to improve estimation precision to some extent by probing into application of assistant information and finery parameters of model.