Learning powerful feature representations for image retrieval has always been a challenging task in the field of remote sensing. Traditional methods focus on extracting low-level hand-crafted features which are not only time-consuming but also tend to achieve unsatisfactory performance due to the complexity of remote sensing images. In this paper, we investigate how to extract deep feature representations based on convolutional neural networks (CNNs) for high-resolution remote sensing image retrieval (HRRSIR). To this end, several effective schemes are proposed to generate powerful feature representations for HRRSIR. In the first scheme, a CNN pre-trained on a different problem is treated as a feature extractor since there are no sufficiently-sized remote sensing datasets to train a CNN from scratch. In the second scheme, we investigate learning features that are specific to our problem by first fine-tuning the pre-trained CNN on a remote sensing dataset and then proposing a novel CNN architecture based on convolutional layers and a three-layer perceptron. The novel CNN has fewer parameters than the pre-trained and fine-tuned CNNs and can learn low dimensional features from limited labelled images. The schemes are evaluated on several challenging, publicly available datasets. The results indicate that the proposed schemes, particularly the novel CNN, achieve state-of-the-art performance.

With the rapid urban development in China, urbanization has brought more and more pressure on the ecological environment. As one of the most dynamic, open, and innovative regions in China, the eco-environmental issues in the Yangtze River Delta have attracted much attention. This paper takes the central region of the Yangtze River Delta as the research object, through building the index system of urbanization and ecological environment based on statistical data and two new indicators (fraction of vegetation coverage and surface urban heat island intensity) extracted from remote sensing images, uses the Entropy-TOPSIS method to complete the comprehensive assessment, and then analyzes the coupling coordination degree between the urbanization and ecological environment and main obstacle factors. The results showed that the coupling coordination degree in the study region generally shows an upward trend from 0.604 in 2008 to 0.753 in 2017, generally changing from an imbalanced state towards a basically balanced state. However, regional imbalance of urbanization and ecological environment always exists, which is mainly affected by social urbanization, economic urbanization, landscape urbanization, pollution loading and resource consumption. Finally, on the basis of the obstacle factor analysis, some specific suggestions for promoting the coordinated development of the Yangtze River Delta are put forward.

The integration of intelligent video surveillance and GIS (geograhical information system) data provides a new opportunity for monitoring and protecting cultivated land. For a GIS-based video monitoring system, the prerequisite is to align the GIS data with video image. However, existing methods or systems have their own shortcomings when implemented in monitoring cultivated land. To address this problem, this paper aims to propose an accurate matching method for projecting vector data into surveillance video, considering the topographic characteristics of cultivated land in plain area. Once an adequate number of control points are identified from 2D (two-dimensional) GIS data and the selected reference video image, the alignment of 2D GIS data and PTZ (pan-tilt-zoom) video frames can be realized by automatic feature matching method. Based on the alignment results, we can easily identify the occurrence of farmland destruction by visually inspecting the image content covering the 2D vector area. Furthermore, a prototype of intelligent surveillance video system for cultivated land is constructed and several experiments are conducted to validate the proposed approach. Experimental results show that the proposed alignment methods can achieve a high accuracy and satisfy the requirements of cultivated land monitoring.

Studies have shown that Sentinel-2 images have advantages over Landsat images in impervious surface area (ISA) extraction. The performance of index-based methods can be affected by different binary methods and subject to seasonal variation. This study marks the first attempt to assess the performance of different spectral indices for ISA extraction using multiseasonal Sentinel-2 images. Specifically, five indices (i.e., the biophysical composition index calculated using the Gram-Schmidt orthogonalization method, biophysical composition index calculated using a principal component-based Procrustes analysis, Normalized Built-up Area Index (NBAI), combinational build-up index, and perpendicular impervious surface index (PISI)) and three impervious surface binary methods (i.e., Otsu's method, manual method, and Iterative Self-Organizing Data Analysis Technique Algorithm (ISODATA) classification method) were tested on multi-seasonal Sentinel-2 images in the main urban area of Wuhan, China. Results showed that PISI combined with the ISODATA classification method achieved the highest accuracy with 92.64% OA, and 0.8410 Kappa coefficient, and NBAI combined with Otsu's method achieved the lowest accuracy with 35.37% OA, and 0.013 Kappa coefficient. Regarding the seasonal sensitivity, PISI is relatively more stable than the other methods. The superior performance of PISI is largely due to its capability in separating ISA from soil and vegetation. In addition, summer is the best season to map ISA from Sentinel-2 images when the impervious surface is generally less confused with bare soil. This study serves as a reference for the selection of spectral indices for ISA extraction from Sentinel-2 images in relation to binary methods and seasonal effects.

Keywords: Cultural Tourism Resources Protection, Building Information Modelling, Indoor-outdoor Seamless Modelling, Four-camera Stereo Photographic Measurement System, 3D Reconstruction, Image-based reconstruction Building the fine 3D model from outdoor to indoor is becoming a necessity for protecting the cultural tourism resources. However, the existing 3D modelling technologies mainly focus on outdoor areas. Actually, a 3D model should contain detailed descriptions of both its appearance and its internal structure, including architectural components. In this paper, a portable four-camera stereo photographic measurement system is developed, which can provide a professional solution for fast 3D data acquisition, processing, integration, reconstruction and visualization. Given a specific scene or object, it can directly collect physical geometric information such as positions, sizes and shapes of an object or a scene, as well as physical property information such as the materials and textures. On the basis of the information, 3D model can be automatically constructed. The system has been applied to the indooroutdoor seamless modelling of distinctive architecture existing in two typical cultural tourism zones, that is, Tibetan and Qiang ethnic minority villages in Sichuan Jiuzhaigou Scenic Area and Tujia ethnic minority villages in Hubei Shennongjia Nature Reserve, providing a new method and platform for protection of minority cultural characteristics, 3D reconstruction and cultural tourism.

Learning powerful feature representations for image retrieval has always been a challenging task in the field of remote sensing. Traditional methods focus on extracting low-level hand-crafted features which are not only time-consuming but also tend to achieve unsatisfactory performance due to the content complexity of remote sensing images. In this paper, we investigate how to extract deep feature representations based on convolutional neural networks (CNN) for high-resolution remote sensing image retrieval (HRRSIR). To this end, two effective schemes are proposed to generate powerful feature representations for HRRSIR. In the first scheme, the deep features are extracted from the fully-connected and convolutional layers of the pre-trained CNN models, respectively; in the second scheme, we propose a novel CNN architecture based on conventional convolution layers and a three-layer perceptron. The novel CNN model is then trained on a large remote sensing dataset to learn low dimensional features. The two schemes are evaluated on several public and challenging datasets, and the results indicate that the proposed schemes and in particular the novel CNN are able to achieve state-of-the-art performance.