Classification of multi‐spectral remote sensing data using a local transfer function classifier

This paper introduces a new neural network, called the local transfer function classifier (LTF‐C), for classification of multi‐spectral remote sensing data. The network structure of LTF‐C is similar to that of the radial basis function neural network (RBF), but LTF‐C utilizes an entirely different learning algorithm. In particular, the network structure of LTF‐C is not predetermined, but changes dynamically during the learning. Such a learning algorithm fits well to the classification problem, and guarantees that the size of the network is as large as is needed. The classification results show that LTF‐C evidently has a better classification accuracy than the six other classifiers in the experiment.     

Bio‐optical model with optimal parameter suitable for Taihu Lake in water colour remote sensing

Sixty‐seven samplings were collected, almost covering all over the Taihu Lake, in one campaign in October 2004. At each station, the backscattering coefficients and the field spectra were measured in situ, respectively, with a HS‐6 and a FieldSpec 931 spectroradiometer (ASD Inc.). Almost concurrently, water samples were fetched with Niskin water‐fetching equipment and then returned to the laboratory for concentration and absorption measurement. The whole lake was divided into different areas according to some indexes. Three models were used to calculate remote sensing reflectance R _rsc for the waters where the in situ remote sensing reflectance R _rsm was beyond the bottom effect, which was considered as optically deep waters. By comparison of R _rsc and R _rsm, the best model suitable for optically deep waters in Taihu Lake, together with its optimal experiential parameter, were selected and developed, which was very important and helpful to develop a universal model to estimate accurately remote sensing reflectance for the whole lake in the next step.     

Analysis of deformation over permafrost regions of Qinghai-Tibet plateau based on permanent scatterers

The surface displacement by seasonally freezing bulge and thawing subsidence are the main hazards for engineering construction in permafrost regions, especially for the Qinghai-Tibet railway. One of the main problems is how to monitor the frozen ground's displacement in the process of construction and protection of the Qinghai-Tibetan railway. The technology of Permanent Scatterers (PS) has been successfully used to detect the long-term subsidence at urban areas. For detecting the subsidence of the frozen earth on Qinghai-Tibet Plateau, this paper extended the capability of the technology of PS to investigate deformation phenomena in vegetated area. The paper analyses an interferometric phase model, and presents improved PS algorithms for separating different components in interferometric phase. The proposed technique is implemented using ENVISAT ASAR images to detect the deformation over permafrost region of Qinghai-Tibet Plateau. The results are in concordance with results provided by a traditional ground levelling, which encourages future development using the Permanent Scatterers method to analyse deformation of the frozen earth on Qinghai-Tibet Plateau.     

Improved sparse representation using adaptive spatial support for effective target detection in hyperspectral imagery

With increasing applications of hyperspectral imagery (HSI) in agriculture, mineralogy, military, and other fields, one of the fundamental tasks is accurate detection of the target of interest. In this article, improved sparse representation approaches using adaptive spatial support are proposed for effective target detection in HSI. For conventional sparse representation, an HSI pixel is represented as a sparse vector whose non-zero entries correspond to the weights of the selected training atoms from a structured dictionary. For improved sparse representation, spatial correlation and spectral similarity of adjacent neighbouring pixels are exploited as spatial support in this context. The size and shape of the spatial support is automatically determined using both adaptive window and adaptive neighbourhood strategies. Accordingly, a solution based on greedy pursuit algorithms is also given to solve the extended optimization problem in recovering the desired sparse representation. Comprehensive experiments on three different data sets using both visual inspection and quantitative evaluation are carried out. The results from these data sets have indicated that the proposed approaches help to generate improved results in terms of efficacy and efficiency.     

Evaluating the performance of PC-ANN for the estimation of rice nitrogen concentration from canopy hyperspectral reflectance

In this study, a wide range of leaf nitrogen concentration levels was established in field-grown rice with the application of three fertilizer levels. Hyperspectral reflectance data of the rice canopy through rice whole growth stages were acquired over the 350 nm to 2500 nm range. Comparisons of prediction power of two statistical methods (linear regression technique (LR) and artificial neural network (ANN)), for rice N estimation (nitrogen concentration, mg nitrogen g^?1 leaf dry weight) were performed using two different input variables (nitrogen sensitive hyperspectral reflectance and principal component scores). The results indicted very good agreement between the observed and the predicted N with all model methods, which was especially true for the PC-ANN model (artificial neural network based on principal component scores), with an RMSE?=?0.347 and REP?=?13.14%. Compared to the LR algorithm, the ANN increased accuracy by lowering the RMSE by 17.6% and 25.8% for models based on spectral reflectance and PCs, respectively.     

A microwave vegetation water index from passive microwave radiometer data

The vegetation water content (VWC) index has been widely used in agriculture, forestry and hydrology studies. It is also useful in retrieving soil moisture from microwave remote sensing observations. Space‐borne and airborne microwave radiometers have widespread utility in soil moisture and vegetation condition retrieval. To simplify the original retrieval algorithm, this paper developed a theoretical microwave vegetation water index (MVWI) from microwave radiometer data, which contains only VWC and a vegetation structure parameter. Based on the MVWI, an efficient VWC retrieval algorithm was developed.     

Can Landsat imagery detect tree line dynamics?

Prior research on tree line dynamics has primarily been based on field inventory data, and little is known about the potential for using remotely sensed imagery to detect change. The present study developed a new methodology by combining remote sensing and field survey data to examine dynamics of the pristine forest in the tree line area on Changbai Mountain in Northeast China over the past two decades. The new method normalized remote sensing data by using the adjacent old‐growth coniferous forest (close to climax) below the tree line as the reference (assuming steady state) to eliminate various potential errors associated with different satellite sensors, atmospheric conditions and seasonal vegetation changes. Specifically, we used a ratio of normalised difference vegetation index (NDVI) between the tree line forest (birch) and the referenced old‐growth coniferous forest, as well as a ratio between the interface forest against the referenced coniferous forest, to investigate growth dynamics of the tree line forest from 1977 to 1999. The interface forest is distributed between the tree line forest and non‐forest zone. Compared with traditional methods using remote sensing data, the new method has higher sensitivity to tree line dynamics. We found that the tree line delineated from satellite imagery had shown no shift in the past decades. However, the NDVI ratio of tree line forest against the reference forest increased from 0.9 to 1.2 from 1977 to 1999, and the ratio of interface forest against the reference forest increased from 0.83 to 0.98, indicating that growth of the tree line forest apparently exceeded that of the reference coniferous forest and had grown denser. Field surveys also supported the conclusion from remote sensing results that the tree line forest on Changbai Mountain has grown denser in the past decades. From 1980 to 2002, basal area of the tree line forest increased 35% (from 18.8 to 25.3 m² ha^?1); at the same time the NDVI ratio increased about 33%. This study suggests that it is possible to monitor growth of the tree line forest based on multi‐temporal remote sensing images. We speculate that global warming may have contributed to the observed rapid growth of the tree line forest because temperature is a major limiting factor among climatic variables in the tree line forest and effect from other factors, such as fire, harvesting and snow and wind damage, were not evident in this protected area in the past decades.     

A knowledge‐based similarity classifier to stratify sample units to improve the estimation precision

This paper presents a comprehensive knowledge‐based similarity classifier that uses remote sensing images and other auxiliary data to map spatial heterogeneity, for stratifying sample units distributed at the geographical landscape in order to improve the precision of the estimate of interest. Our method emphasises the decrease of bias so as to produce the high‐quality stratifying frame. For this purpose, the method takes some necessary measures such as use of auxiliary variables including spectral bands, physical and socioeconomic data to help cluster analysis, correlation analysis between auxiliary variables and the goal variable to remove irrelevant data and consideration of spatial correlation in cluster analysis through the density‐based unsupervised learning etc. Furthermore, considering the time‐consuming characteristic of clustering huge spatiotemporal datasets, the method uses non‐parametric supervised learning to induce rules for clustered classes. The rules could be efficiently used to group pixels into different classes of similarity. Then in the method, the pixel‐level similarity image was vectorised into polygons with different group labels, thus producing the vector map of geospatial heterogeneity as an easy‐to‐use stratification frame. Last, to have an accurate estimation of the goal variable, our method re‐divided sample units while the units covered by different strata and considered the effect of the sample size in the estimation algorithm. In the survey case of the cultivated land area, the proposed method achieves higher accuracy and a better coefficient of relative efficiency (RE) of stratification with its estimate closer to the observed value in comparison with other stratification strategies, e.g., k‐means, SOM and those similar to eco‐regions. Our method has potential practical merits as a good stratification strategy can increase the precision and considerably save the cost of sampling for many large regions, such as those in China, to be surveyed.     

Side‐scan sonographs of steep slopes in the Wushieh Reservoir

Operators tow side‐scan sonar transducers at the optimum height from the bottom to produce better quality side‐scan sonographs. This can be done only if the bottom is moderately flat. Sonographs with maximum tonal intensity are easily produced on steep slopes e.g. valley banks. Strong signatures are recorded over all ranges due to the poor geometry of sonographs produced without acoustic shadows. In order to create clear scanned images of a steep slope using a side‐scan sonar, rotation of the towed fish or tilting the transducer face to the dipping slope surface is suggested. Cases conducted on the shore of the Wushieh Reservoir, Taiwan, in which the bank slopes varied from 10° to 45° are discussed in this work. In general, the angle of the scanned slope determines the degree of towed fish rotation angle. The rotation angle is set at a smaller angle than the slope, giving stronger returns from the lower slope. In contrast, if the rotation angle is set at a larger angle a null return outside the insonified area in the lower slope is produced. A sub‐aqueous landslide was observed in the side‐scan sonographs using this strategy. This study also gives results that affirm the value of a side‐scan sonar as a powerful tool for underwater inspection.     

Shadow information recovery in urban areas from very high resolution satellite imagery

It is usually quite difficult to extract and recover shadow information in the urban environment from remote sensing imagery. This paper describes the study of precisely detected shadow in satellite images and recovering information from the surface covered in shadow from very high resolution (VHR) satellite imagery.