Object-based image analysis (OBIA) is regarded as an effective technology for high-spatial resolution (HSR) image classification due to its clear and intuitive technical process. However, OBIA depends on the manual tuning of image classification features, which is a tricky job. Deep learning (DL) technology auto-learns image features from massive images and obtains higher image classification accuracy than traditional techniques. In this study, a novel method called object-scale adaptive convolutional neural network (OSA-CNN), which combines OBIA with CNN, is proposed for HSR image classification. First, OSA-CNN collects the image patches along the main axes of the object primitives obtained through image segmentation; the size of the former is automatically determined through the axis widths of the latter. This step generates the input units required for CNN classification. Second, the Squeeze-and-Excitation block is extracted from the SE network into the network structure of GoogleNet, which realizes the weighted fusion of the multiscale convolutional features, enhances useful features, and suppresses useless ones. In the classification stage, multiscale image segmentation and CNN classification are fused using an object-scale adaptive mechanism. Finally, object primitives are classified through majority voting on the image patches. The network structure modifications, multiscale classification fusion, and other improvements are verified by gradually incorporating these steps into the original GoogleNet. The experiments show that these improvements effectively enhance the image classification accuracy. This study presents an effective way of combining OBIA and DL techniques to utilize the advantages of both approaches and facilitate HSR image classification.

To protect and restore this downstream ecosystem, the Tarim River Basin Administration Bureau (TBAB) initiated the Ecological Water Compensation (EWC) project from 2000 to 2018. Revealing the mechanism of vegetation-hydroecological response processes in the lower reaches of the Tarim River before and after EWC work is conducive to water resource planning, utilization and protection. In this paper, the spatiotemporal responses of vegetation and groundwater to EWC were examined at the points, lines, and area (PLA) scale by coupling remote sensing techniques and field station observation data collected between 2000 and 2017. The findings indicated that (1) In general, the regional fractional vegetation coverage (FVC) increased significantly, and the average FVC growth rate was 3.5%/year from 2000 to 2017 (R² > 0.84, p < 0.01, 2-tailed). (2) The regional vegetation restoration process showed obvious fluctuations and stage characteristics, but the spatial scope of the significantly increased vegetation area was limited. Plants grew rapidly within 10 km of the river, while 10 km away from the water channel, no obvious change was observed. (3) Strong coupling relationships were identified among the FVC growth, EWC volume and groundwater depth variations (p < 0.05, 2-tailed). The response times of the regional vegetation and groundwater depth to EWC indicated one-year lags. The above results imply that the regional ecological environment was significantly improved over the study period, thus confirming that the EWC project made remarkable accomplishments. However, the effect of ecological restoration is not sufficiently stable at present. Vegetation restoration has mainly been centralized around the river channel and is greatly dependent on the annual EWC volume. In addition, the local conditions begin to degrade soon after an EWC project is terminated, and vice versa; when EWC commences, the FVC immediately begins to improve. Therefore, the current EWC achievements need to be further consolidated and strengthened in the future.

Recent building-oriented change detection studies considered only morphological changes in buildings, and few publicly available change detection datasets further distinguish among building types. In this study, we propose a semantic change detection network named BSCDNet that considers both morphological and semantic changes in buildings. BSCDNet adopts a multitasking branch structure with object classification, change detection and segmentation to simultaneously achieve object-level semantic classification and change analysis of buildings in bitemporal, high-spatial-resolution (HSR) imagery. In the object classification branch, a graph attention network (GAT) is utilized to capture the spatial and semantic correlations among buildings during classification. The change detection branch applies both spatial and channel attention mechanisms to eliminate nonbuilding interference and enhance change features. Moreover, the segmentation branch adopts a distinctive instance segmentation procedure that improves the accuracy of object segmentation. We created a building change detection dataset with category differentiation based on HSR imagery to validate the proposed method. Ablation experiments verify the effectiveness and advantages of the above task branches. Furthermore, in comparative experiments with several SOTA semantic change detection methods, such as HRSCD, SCDNet and MR_CD, BSCDNet reached the optimal level in terms of F1 and mIoU when evaluating the change detection performance, as well as kappa and score for evaluating the classification performance.

To address various frequency features differentiated modeling in complex scenes during remote sensing image super-resolution reconstruction, this paper proposes a novel method that integrates a wavelet query multi-head attention mechanism to enhance the super- resolution reconstruction of remote sensing images. We construct an enhanced generative adversarial network based on this mechanism, termed Wavelet Query Multi-Attention Generative Adversarial Network (WQMAGAN). In the generator, wavelet-query frequency decomposition is utilized to achieve differentiated modeling of features across various frequency bands. Meanwhile, a channel attention mechanism is incorporated to capture global inter-channel dependencies, thereby enhancing the reconstruction capability for high-frequency textures and edge structures. In the discriminator stage, the reconstruction scene discriminability is improved by embedding an Omni-Dimensional Dynamic Convolution (ODConv) based attention mechanism within the U-Net architecture. This mechanism adaptively modulates detailed texture features, providing more stable and discriminative structural and textural constraints for the generated results. Experimental results demonstrate that WQMAGAN achieves a balanced reconstruction of both high-frequency textures and low-frequency structures. The learned perceptual image patch similarity (LPIPS) reaches a low value of 0.360, the peak signal-to-noise ratio (PSNR) achieves 29.31, and the structural similarity index measure (SSIM) reaches 0.692, while the no-reference quantitative evaluation metrics perception- based image quality evaluator (PIQE) and naturalness image quality evaluator (NIQE) outperform existing state-of-the-art image super-resolution methods. WQMAGAN exhibits excellent performance and strong robustness, significantly improving the reconstruction quality of remote sensing images and showcasing broad market prospects. Our code and pre-trained models will be available on the GitHub platform (https://github.com/BEABOSS/WQMAGAN).