Remote Sensing Image Classification Based on Improved Fuzzy c-Means

Classification is always the key point in the field of remote sensing. Fuzzy c-Means is a traditional clustering algorithm that has been widely used in fuzzy clustering. However, this algorithm usually has some weaknesses, such as the problems of falling into a local minimum, and it needs much time to accomplish the classification for a large number of data. In order to overcome these shortcomings and increase the classifi-cation accuracy, Gustafson-Kessel (GK) and Gath-Geva (GG) algorithms are proposed to improve the tradi-tional FCM algorithm which adopts Euclidean distance norm in this paper. The experimental result shows that these two methods are able to detect clusters of varying shapes, sizes and densities which FCM cannot do. Moreover, they can improve the classification accuracy of remote sensing images.

Neural Network Based on Rough Sets and Its Application to Remote Sensing Image Classification

This paper presents a new kind of back propagation neural network (BPNN) based on rough sets,called rough back propagation neural network (RBPNN).The architecture and training method of RBPNN are presented and the survey and analysis of RBPNN for the classification of remote sensing multi_spectral image is discussed.The successful application of RBPNN to a land cover classification illustrates the simple computation and high accuracy of the new neural network and the flexibility and practicality of this new approach.

A RBF classification method of remote sensing image based on genetic algorithm

The remote sensing image classification has stimulated considerable interest as an effective method for better retrieving information from the rapidly increasing large volume, complex and distributed satellite remote imaging data of large scale and cross-time, due to the increase of remote image quantities and image resolutions. In the paper, the genetic algorithms were employed to solve the weighting of the radial basis faction networks in order to improve the precision of remote sensing image classification. The remote sensing image classification was also introduced for the GIS spatial analysis and the spatial online analytical processing (OLAP), and the resulted effectiveness was demonstrated in the analysis of land utilization variation of Daqing city.

Classification Method Research to Remote Sensing Images

With rapid development of remote sensing technology, the resolution of remote sensing images is increasingly improved; then people can extract more useful data and information from these images. Thus, an important information extraction method from remote sensing images - image classification, becomes more and more important. Based on phenopthase and band composition characteristics, this paper firstly discusses the important role of background parameters in remote sensing images classification; then based on geographical infomation system technology, the computerized automatic classification to high-medium-low-yield croplands in Dingxiang County of Shanxi Province in rotate sensing images has been carried out by using eompound layers classification method of multi-thematic information; compared the classification result to the visual interpretation results, the accuracy increases from 70% to above 90%.

Semi-supervised kernel FCM algorithm for remote sensing image classification

These problems of nonlinearity, fuzziness and few labeled data were rarely considered in traditional remote sensing image classification. A semi-supervised kernel fuzzy C-means （SSKFCM） algorithm is proposed to overcome these disadvantages of remote sensing image classification in this paper. The SSKFCM algorithm is achieved by introducing a kernel method and semi-supervised learning technique into the standard fuzzy C-means （FCM） algorithm. A set of Beijing-1 micro-satellite＇s multispectral images are adopted to be classified by several algorithms, such as FCM, kernel FCM （KFCM）, semi-supervised FCM （SSFCM） and SSKFCM. The classification results are estimated by corresponding indexes. The results indicate that the SSKFCM algorithm significantly improves the classification accuracy of remote sensing images compared with the others.

Identification and spatial pattern analysis of alpine timberline by remote sensing methods in Yarlung Zangpo Grand Canyon

As an important ecotone,the alpine timberline is the boundary between closed-canopy montane forest and alpine vegetation,and is highly sensitive to global and regional climate changes.We provided a way to identify and extract the alpine timberline in Yarlung Zangpo Grand Canyon Nature Reserve by using remote sensing data and spatial analysis based on land use/land cover classification and NDVI distribution characteristics.Combining DEM data,the influence of slope and aspect on the distribution of alpine timberline was explored.The results showed that the alpine timberline in Yarlung Zangpo Grand Canyon is transitional timberline,with the upper boundary approximately distributed at the elevation of 3422-4373 m,the lower boundary at approximately 3270-4164 m,with a width of about 110-280 m.Alpine timberline was mainly distributed on steep and very steep slopes ranging from 25°to 45°.The maximum elevation of both the upper and lower boundaries occurred on steep slopes.The distribution of alpine timberline varies with aspects,with sunny slopes having a higher boundary than shady slopes.

A new ensemble feature selection and its application to pattern classification

Neural network ensemble based on rough sets reduct is proposed to decrease the computational complexity of conventional ensemble feature selection algorithm. First, a dynamic reduction technology combining genetic algorithm with resampling method is adopted to obtain reducts with good generalization ability. Second, Multiple BP neural networks based on different reducts are built as base classifiers. According to the idea of selective ensemble, the neural network ensemble with best generalization ability can be found by search strategies. Finally, classification based on neural network ensemble is implemented by combining the predictions of component networks with voting. The method has been verified in the experiment of remote sensing image and five UCI datasets classification. Compared with conventional ensemble feature selection algorithms, it costs less time and lower computing complexity, and the classification accuracy is satisfactory.

Research on Extraction Method of Surface Information Based on Multi-Feature Combination Such as Fractal Texture

Because of the developed economy and lush vegetation in southern China, the following obstacles or difficulties exist in remote sensing land surface classification: 1) Diverse surface composition types;2) Undulating terrains;3) Small fragmented land;4) Indistinguishable shadows of surface objects. It is our top priority to clarify how to use the concept of big data (Data mining technology) and various new technologies and methods to make complex surface remote sensing information extraction technology develop in the direction of automation, refinement and intelligence. In order to achieve the above research objectives, the paper takes the Gaofen-2 satellite data produced in China as the data source, and takes the complex surface remote sensing information extraction technology as the research object, and intelligently analyzes the remote sensing information of complex surface on the basis of completing the data collection and preprocessing. The specific extraction methods are as follows: 1) extraction research on fractal texture features of Brownian motion;2) extraction research on color features;3) extraction research on vegetation index;4) research on vectors and corresponding classification. In this paper, fractal texture features, color features, vegetation features and spectral features of remote sensing images are combined to form a combination feature vector, which improves the dimension of features, and the feature vector improves the difference of remote sensing features, and it is more conducive to the classification of remote sensing features, and thus it improves the classification accuracy of remote sensing images. It is suitable for remote sensing information extraction of complex surface in southern China. This method can be extended to complex surface area in the future.

Efficient knowledge distillation for hybrid models:A vision transformer‐convolutional neural network to convolutional neural network approach for classifying remote sensing images

In various fields,knowledge distillation(KD)techniques that combine vision transformers(ViTs)and convolutional neural networks(CNNs)as a hybrid teacher have shown remarkable results in classification.However,in the realm of remote sensing images(RSIs),existing KD research studies are not only scarce but also lack competitiveness.This issue significantly impedes the deployment of the notable advantages of ViTs and CNNs.To tackle this,the authors introduce a novel hybrid‐model KD approach named HMKD‐Net,which comprises a CNN‐ViT ensemble teacher and a CNN student.Contrary to popular opinion,the authors posit that the sparsity in RSI data distribution limits the effectiveness and efficiency of hybrid‐model knowledge transfer.As a solution,a simple yet innovative method to handle variances during the KD phase is suggested,leading to substantial enhancements in the effectiveness and efficiency of hybrid knowledge transfer.The authors assessed the performance of HMKD‐Net on three RSI datasets.The findings indicate that HMKD‐Net significantly outperforms other cuttingedge methods while maintaining a significantly smaller size.Specifically,HMKD‐Net exceeds other KD‐based methods with a maximum accuracy improvement of 22.8%across various datasets.As ablation experiments indicated,HMKD‐Net has cut down on time expenses by about 80%in the KD process.This research study validates that the hybrid‐model KD technique can be more effective and efficient if the data distribution sparsity in RSIs is well handled.

Remote Sensing Classification of Marsh Wetland with Different Resolution Images

Successful biological monitoring depends on judicious classification. An attempt has been made to provide an overview of important characteristics of marsh wetland. Classification was used to describe ecosystems and land cover patterns. Different spatial resolution images show different landscape characteristics. Several classification images were used to map and monitor wetland ecosystems of Honghe National Nature Reserve （HNNR） at a plant community scale. HNNR is a typical inland wetland and fresh water ecosystem in the North Temperate Zone. SPOT-5 10 m ×10 m, 20 m × 20 m, and 30 m×30 m images and Landsat -5 Thematic Mapper （TM） images were used to classify based on maximum likelihood classification （MLC） algorithms. In order to validate the precision of the classifications, this study used aerial photography classification maps as training samples because of their high accuracy. The accuracy of the derived classes was assessed with the discrete multivariate technique called KAPPA accuracy. The results indicate： （1） training samples are important to classification results. （2） Image classification accuracy is always affected by areal fraction and aggregation degree as well as by diversities and patch shape. （3） The core zone area is protected better than buffer zone and experimental zone wetland. The experimental zone degrades fast because of irrational development by humans.

Remote sensing-based artificial surface cover classification in Asia and spatial pattern analysis

Artificial surfaces, characterized with intensive land-use changes and complex landscape structures, are important indicators of human impacts on terrestrial ecosystems. Without high-resolution land-cover data at continental scale, it is hard to evaluate the impacts of urbanization on regional climate, ecosystem processes and global environment. This study constructed a hierarchical classification system for artificial surfaces, promoted a remote sensing method to retrieve subpixel components of artificial surfaces from 30-m resolution satellite imageries(Globe Land30) and developed a series of data products of high-precision urban built-up areas including impervious surface and vegetation cover in Asia in 2010. Our assessment, based on multisource data and expert knowledge, showed that the overall accuracy of classification was 90.79%. The mean relative error for the impervious surface components of cities was 0.87. The local error of the extracted information was closely related to the heterogeneity of urban buildings and vegetation in different climate zones. According to our results, the urban built-up area was 18.18×104 km2, accounting for 0.59% of the total land surface areas in Asia; urban impervious surfaces were 11.65×104 km2, accounting for 64.09% of the total urban built-up area in Asia. Vegetation and bare soils accounted for 34.56% of the urban built-up areas. There were three gradients: a concentrated distribution, a scattered distribution and an indeterminate distribution from east to west in terms of spatial pattern of urban impervious surfaces. China, India and Japan ranked as the top three countries with the largest impervious surface areas, which respectively accounted for 32.77%, 16.10% and 11.93% of the urban impervious surface area of Asia. We found the proportions of impervious surface and vegetation cover within urban built-up areas were closely related to the economic development degree of the country and regional climate environment. Built-up areas in developed countries had relatively low impervious surface and high public green vegetation cover, with 50–60% urban impervious surfaces in Japan, South Korea and Singapore. In comparison, the proportion of urban impervious surfaces in developing countries is approaching or exceeding 80% in Asia. In general, the composition and spatial patterns of built-up areas reflected population aggregation and economic development level as well as their impacts on the health of the environment in the sub-watershed.

Classification of Remote Sensing Images Based on Band Selection and Multi-mode Feature Fusion

As feature data in multimodal remote sensing images belong to multiple modes and are complementary to each other,the traditional method of single-mode data analysis and processing cannot effectively fuse the data of different modes and express the correlation between different modes.In order to solve this problem,make better fusion of different modal data and the relationship between the said features,this paper proposes a fusion method of multiple modal spectral characteristics and radar remote sensing imageaccording to the spatial dimension in the form of a vector or matrix for effective integration,by training the SVM model.Experimental results show that the method based on band selection and multi-mode feature fusion can effectively improve the robustness of remote sensing image features.Compared with other methods,the fusion method can achieve higher classification accuracy and better classification effect.

How to accurately extract large-scale urban land?Establishment of an improved fully convolutional neural network model

Realizing accurate perception of urban boundary changes is conducive to the formulation of regional development planning and researches of urban sustainable development.In this paper,an improved fully convolution neural network was provided for perceiving large-scale urban change,by modifying network structure and updating network strategy to extract richer feature information,and to meet the requirement of urban construction land extraction under the background of large-scale low-resolution image.This paper takes the Yangtze River Economic Belt of China as an empirical object to verify the practicability of the network,the results show the extraction results of the improved fully convolutional neural network model reached a precision of kappa coefficient of 0.88,which is better than traditional fully convolutional neural networks,it performs well in the construction land extraction at the scale of small and medium-sized cities.