Building extraction based on remote sensing can provide reliable geographic basic information and used in many fields. Deep learning has strong feature mining capabilities and is adept at solving image processing related problems such as image classification and object detection, thus having great potential in the field of building extraction. U-Net, as a deep convolutional neural network used for image segmentation tasks, can achieve precise pixel level segmentation. However, due to the limitations of its network structure, U-Net has a slight lack of accuracy in extracting buildings with small sizes, complex or fuzzy boundaries, and complex spatial distribution. For this reason, this study improves the U-Net network in these aspects: (1) The layers of the U-Net model are deepened for enhancing the feature extraction ability; (2) Fully convolutional network (FCN) decoder is introduced as an auxiliary loss function module to improve the efficiency and effect of the model training; (3) Elu activation function is introduced to improve the efficiency of model back propagation; and (4) Dice loss function is introduced to solve the problem of data imbalance. Compared to the original U-Net, the improved U-Net has better building extraction performance under complex spatial distribution and contours, with improvements of about 18.40%, 20.61%, and 19.69% in accuracy, recall, and F1 values, respectively.

In some scenes with complex terrain or features, the accuracy of the constructed three-dimensional (3D) based on oblique photogrammetry model is not ideal, and it often requires manual placement of ground control points (GCPs). What is the impact of GCPs on the accuracy of the constructed 3D model? This study designs a detailed experiment to explore this issue by collecting unmanned aerial vehicle (UAV) images and coordinates of control points in the testing area, adding different numbers of GCPs for 3D modeling, and finally conducting a systematic analysis of the accuracy of the 3D realistic model of the component from both qualitative and quantitative perspectives. The experiment shows that the effect of adding GCPs on the 3D model has been greatly improved, with the error in plane accuracy reduced to 1/14 of that without control points, and the horizontal and vertical deformations also decreased by about 10%. As the number of GCPs increases, the impact of enhancing 3D modeling precision will gradually diminish.

The semantic segmentation of vegetation point clouds has very important application value in the field of geosciences. It can distinguish vegetation regions from other regions, further classify and analyze the vegetation, and help us better understand the distribution and characteristics of vegetation to protect and manage natural resources. The PointNet and PointNet++ models use maximum pooling as the aggregation function, allowing the deep neural networks to classify unordered point clouds directly with high classification accuracy. However, their ability to extract spatial correlations and local features from point clouds is insufficient, which restricts the improvement of point clouds semantic segmentation accuracy and results in the poor processing of vegetation point clouds. To resolve this problem, this research designs the novel hierarchical point cloud transformer (HPCT) model, suitable for the semantic segmentation of multisource vegetation point clouds. Combined with deep learning techniques, different levels of features are processed hierarchically based on a hierarchical structure, and a Transformer module is combined in the feature extraction part, so as to obtain a larger receptive field and stronger semantic feature extraction capability. At the same time, we also propose a unified spatial scale sampling method for heterogeneous point cloud data input, which can be used not only for training and predicting the independent HPCT models with a single source of data, but also for training and predicting a unified HPCT model with multisource data. Semantic segmentation experiments are carried out on self-collected three-source data sets. The results show that the semantic segmentation performance of evaluation indicators (such asRecall,Pre,IoU, andOA) of the proposed HPCT model under the independent training and unified training on the three-source data exceed those of the PointNet, PointNet++, and PCT models, and even exceed some newly emerging models, such as PontCNN and DGCNN. The unified HPCT model has better segmentation performance than the independent HPCT model, with averageRecall,Pre,IoU, andOAindicators increasing by 1.07%, 1.73%, 4.33%, and 1.03%, respectively. We attribute this superior accuracy to the unified training with the three-source data. The averageRecall,Pre,IoU, andOAindicators of the unified HPCT model for the entire three-source data set exceed 96%, 98%, 95%, and 98%, respectively.

In this paper, a novel interference mitigation approach using an autoencoder in combination with a traditional interference detection filter is introduced. It is shown that by employing the gated convolution, the encoder has the ability to learn the signal pattern from the remaining interference-free signal. The decoder can recover the interference-contaminated signal segments from the bottleneck representation as computed by the encoder. Experimental results show that the proposed method can provide a remarkable improvement in signal-to-interference-plus-noise ratio (SINR) and preserves its robustness on real radar measurements in severely disturbed scenarios that are more complex than the training dataset.

This letter mainly deals with the problem of counting moving human targets in an enclosed building space for through-the-wall radar. Specifically, a typical deep convolutional neural network, namely, residual neural network (ResNet), is designed to identify the line-like texture information associated with the target number from the blurred range-time images of a single-channel stepped-frequency continuous-wave (SFCW) radar. Experiments demonstrate that the ResNet-based counting algorithm achieves an accuracy of 91.54% for one to six human targets, and the accuracy rises to 97.12% when only counting one to three humans, even under conditions of wall penetration degradation, limited spatial resolution, heavy multipath clutters, and target-to-target occlusion. The achieved number of information of moving human targets not only contributes directly to the situation assessment behind the wall but also can act as the prior information to promote further target detection.

When dealing with salient object that contains several regions with different appearances, salient object detection can be a difficult task as often only parts of the salient object are highlighted and consistency between the salient regions is poor. This study tackles this problem by introducing objectness to assist the salient object detection. Rather than treating objectness in the same manner as other low‐level cues (e.g. uniqueness, location etc.) for the determination of regional saliency values, the authors emphasise that objectness should also play a significant role in tuning the consistency between salient regions. The authors integrate objectness, uniqueness and centre bias to find potential salient regions and then enforce consistency between these regions using a full‐connected Gaussian Markov random field with the weights determined by the objectness score. Experimental results on public benchmark datasets indicate that the authors’ method performs well on many images which cannot be well detected traditionally.