The automatic extraction of building outlines from high-resolution images is an important and challenging task. Convolutional neural networks have shown excellent results compared with traditional building extraction methods because of their ability to extract high-level abstract features from images. However, it is difficult to fully utilize the multiple features of current building extraction methods; consequently, the resulting building boundaries are irregular. To overcome these limitations, we propose a method for extracting buildings from high-resolution images using a multifeature convolutional neural network (MFCNN) and morphological filtering. Our method consists of two steps. First, the MFCNN, which consists of residual connected unit (RCU), dilated perception unit (DPU), and pyramid aggregation unit (PAU), is used to achieve pixel-level segmentation of the buildings. Second, morphological filtering is used to optimize the building boundaries, improve the boundary regularity, and obtain refined building boundaries. The Massachusetts and Inria datasets are selected for experimental analysis. Under the same experimental conditions, the extraction results achieved with the proposed MFCNN are compared with the results of other deep learning models that have been commonly used in recent years: FCN-8s, SegNet, and U-Net. The results on both datasets reveal that the proposed model improves the F1-score by 3.31%-5.99%, increases the overall accuracy (OA) by 1.85%-3.07%, and increases the intersection over union (IOU) by 3.47%-8.82%. These findings illustrate that the proposed method is effective at extracting buildings from complex scenes.

Since fire is one of the most serious types of accidents that can occur, there is always a need for improvement in fire detection capabilities. Convolutional neural networks (CNNs) have been used for a variety of high-performance computer vision tasks. The use of CNNs to extract deep static features of fire has greatly improved the accuracy of fire detection. However, the implementation of CNNs in the real world is limited by their high computational cost. In addition, fire detection methods based on the classification of images alone using CNNs cannot account for the dynamic features of fire. Therefore, in this paper, a method that exploits both motion-flicker-based dynamic features and deep static features is proposed for video fire detection. First, dynamic features are extracted by analyzing the differences in motion and flicker features between fire and other objects in videos. Second, an adaptive lightweight convolutional neural network (ALCNN) is proposed to extract the deep static features of fire. Finally, the dynamic and static features of fire are combined to establish a video fire detection method with improved operational efficiency in terms of accuracy and run time. To prove the validity of our method, its accuracy and run time are evaluated on three test datasets, and the results reveal that our method exhibits better performance than state-of-the-art methods. Moreover, our method is shown to be feasible in complex video scenarios and for devices with resource constraints.

Accurately building height estimation from remote sensing imagery is an important and challenging task. However, the existing shadow-based building height estimation methods have large errors due to the complex environment in remote sensing imagery. In this paper, we propose a multi-scene building height estimation method based on shadow in high resolution imagery. First, the shadow of building is classified and described by analyzing the features of building shadow in remote sensing imagery. Second, a variety of shadow-based building height estimation models is established in different scenes. In addition, a method of shadow regularization extraction is proposed, which can solve the problem of mutual adhesion shadows in dense building areas effectively. Finally, we propose a method for shadow length calculation combines with the fish net and the pauta criterion, which means that the large error caused by the complex shape of building shadow can be avoided. Multi-scene areas are selected for experimental analysis to prove the validity of our method. The experiment results show that the accuracy rate is as high as 96% within 2 m of absolute error of our method. In addition, we compared our proposed approach with the existing methods, and the results show that the absolute error of our method are reduced by 1.24 m–3.76 m, which can achieve high-precision estimation of building height.

Vision transformers (ViTs) are increasingly utilized for HSI classification due to their outstanding performance. However, ViTs encounter challenges in capturing global dependencies among objects of varying sizes, and fail to effectively exploit the spatial–spectral information inherent in HSI. In response to this limitation, we propose a novel solution: the multi-scale spatial–spectral transformer (MSST). Within the MSST framework, we introduce a spatial–spectral token generator (SSTG) and a token fusion self-attention (TFSA) module. Serving as the feature extractor for the MSST, the SSTG incorporates a dual-branch multi-dimensional convolutional structure, enabling the extraction of semantic characteristics that encompass spatial–spectral information from HSI and subsequently tokenizing them. TFSA is a multi-head attention module with the ability to encode attention to features across various scales. We integrated TFSA with cross-covariance attention (CCA) to construct the transformer encoder (TE) for the MSST. Utilizing this TE to perform attention modeling on tokens derived from the SSTG, the network effectively simulates global dependencies among multi-scale features in the data, concurrently making optimal use of spatial–spectral information in HSI. Finally, the output of the TE is fed into a linear mapping layer to obtain the classification results. Experiments conducted on three popular public datasets demonstrate that the MSST method achieved higher classification accuracy compared to state-of-the-art (SOTA) methods.

Extracting building heights from single-view remote sensing images greatly enhances the application of remote sensing data. While methods for extracting building height from single-view shadow images have been widely studied, it remains a challenging task. The main reasons are as follows: (1) The traditional method for extracting shadow information exhibits low accuracy. (2) The use of only shadow information to extract building height results in limited application scenarios. To solve the above problems, this paper introduces building side and shadow information to complement each other, and proposes a building height extraction method from high-resolution single-view remote sensing images using shadow and side information. Firstly, we propose the RMU-Net method, which utilizes multi-scale features for the extraction of shadow and side information. This method aims to address issues related to pixel detail loss and imprecise edge segmentation, which result from significant scale differences within segmentation targets. Additionally, we employ the area threshold method to optimize the segmentation results, specifically to tackle small stray patches and holes, enhancing the overall integrity and accuracy of shadow and side information extraction. Secondly, we propose a method for building height extraction that integrates shadow and side information based on an enhanced proportional coefficient model. The accuracy of measuring building side and shadow lengths is improved by incorporating the fishing net method, informed by our analysis of the geometric relationships among buildings. Finally, we establish a dataset containing building shadow and side information from remote sensing images, and select multiple areas for experimental analysis. The results demonstrate a shadow extraction accuracy of 91.03% and a side extraction accuracy of 90.29%. Additionally, the average absolute error (MAE) for building height extraction is 1.22, while the average root mean square error (RMSE) is 1.21. Furthermore, the proposed method's validity and scalability are affirmed through experimental analyses of applicability and anti-interference performance in extensive areas.