For spatially constrained scenarios where the target is not within the camera's field of view, mirror-based imaging techniques are widely used for camera pose estimation. However, in industrial environments, mechanical vibrations introduce dynamic perturbations, causing small real-time displacements of the camera, mirror, and target, which increase the difficulty of pose estimation. Effective solutions under such conditions remain limited. To address this issue, we propose a camera pose estimation method based on binocular pose relationships under mirror camera constraints. The method employs a rigidly coupled camera-mirror configuration and treats the mirror pose in the camera coordinate system as a stable intermediate representation to mitigate the effects of vibration-induced perturbations. Geometric constraints derived from binocular pose relationships are incorporated into a least-squares framework to obtain an initial estimate of the mirror pose. Building upon this, we propose a dual-stage optimization strategy to progressively refine the mirror pose, thereby improving estimation accuracy and robustness. Finally, the camera pose is recovered from the stabilized mirror pose using conventional pose estimation methods. The accuracy and stability of the proposed method are validated through both simulation and real-world experiments. In simulation experiments with pixel-level noise, the average angular error of the mirror normal vector is 0.007 rad, and the average distance error is 0.860 mm. In real-world experiments, the standard deviations of the translation of the mirror pose are 0.919 mm and 1.184 mm in two experimental groups, both of which are smaller than those of the comparison methods. Furthermore, the proposed method achieves the lowest reprojection errors for camera pose estimation in both groups, with reductions of approximately 22%-87% and 16%-82% compared to the other methods.

Scarce feature points are a critical limitation affecting the accuracy and stability of incremental structure from motion (SfM) in small-scale scenes. In this paper, we propose an incremental SfM method for small-scale scenes, combined with an auxiliary calibration plate. This approach increases the number of feature points in sparse regions, and we randomly generate feature points within those areas. At the same time, we obtain a coarse matching set of feature points using pairwise polar geometric constraints. The positional results from the geometric constraints of the calibration plate are then used to filter out high-precision matching points, thereby improving the accuracy of the three-dimensional reconstruction. Experimental results demonstrate that the proposed method achieves superior reconstruction completeness and accuracy. In three real-world experiments, the average re-projection errors were 0.5245, 0.4151, and 0.4996 pixels, outperforming competing methods. This approach ensures robust pose estimation and facilitates precise 3D reconstructions.

Objective Thin cloud contamination in remote sensing images presents a significant challenge affecting data quality,resulting in imprecise analysis and interpretation across applications including land cover classification,environmental monitoring,and disaster assessment.Conventional thin cloud removal methods typically depend on feature extraction at a single scale and inadequately capture the multi-scale characteristics of clouds,leading to suboptimal declouding results.Furthermore,deep learning-based approaches,particularly those utilizing generative adversarial network(GAN),frequently encounter detail loss and texture blur in generated images and demonstrate limited capability in modeling local features accurately.To address these challenges,this study introduces a novel GAN-based method incorporating a convolutional block attention module(CBAM)and a multi-scale attention mechanism.The proposed approach aims to enhance the accuracy of thin cloud removal while maintaining the spectral and spatial details of the original imagery,thus improving the overall quality of remote sensing data. Methods The proposed framework integrates the GAN architecture with CBAM and multi-scale attention mechanism for effective thin cloud removal.The generator network is engineered to capture global and local features of the input image,enabling the model to restore detailed surface information while removing thin clouds effectively.The discriminator network assesses the authenticity of the generated image,ensuring high similarity to the real cloud-free image.The multi-scale attention mechanism serves a crucial function by implementing parallel convolution branches with independent parameter optimization strategies.This approach enables differentiated feature expression,enhancing the model's capacity to process cloud contamination and underlying surface features at various scales.Furthermore,CBAM is integrated for enhanced feature extraction at different scales.CBAM applies sequential channel and spatial attention to feature maps,adaptively emphasizing important features while suppressing irrelevant noise.This integration of multi-scale attention and CBAM substantially improves the model's capability to restore image brightness and recover fine details.Comprehensive experiments were conducted on the RICE1 dataset and a custom remote sensing cloud removal dataset based on Sentinel-2 imagery.The model's performance is evaluated using quantitative metrics including peak signal-to-noise ratio(PSNR)and structural similarity index(SSIM).The proposed method is compared against several state-of-the-art thin cloud removal techniques,including Haze Removal,FFA-Net,C2PNet,CGAN,and SpA-GAN,to demonstrate its effectiveness. Results and Discussions Experimental results show that the proposed method surpasses traditional thin cloud removal techniques in both visual quality and quantitative metrics.The integration of CBAM with the multi-scale attention mechanism substantially enhances the model's ability to recover detailed surface information while effectively removing thin clouds(Figs.8-11).Comparative analysis reveals that the proposed method achieves a PSNR of 31.321 dB and an SSIM of 0.894,exceeding the performance of state-of-the-art methods(Tables 1 and 2).The generated images are further analyzed based on the average brightness of the RGB channels(Figs.11 and 12).The results indicate that the cloud-free images generated by the proposed method most closely match the real images in terms of RGB channel brightness,validating the method's effectiveness in preserving spectral details.An ablation study examines the synergistic contribution of the two attention mechanisms(Table 4 and Fig.14).The results confirm that their combination significantly enhances model performance,demonstrating their complementary role in improving image quality.Specifically,the multi-scale attention mechanism facilitates feature capture at different scales,while CBAM enhances feature extraction accuracy through channel and spatial dimension focus. Conclusions This study presents a novel method for thin cloud removal from remote sensing images based on GAN enhanced with CBAM and multi-scale attention mechanism.The proposed approach enhances cloud removal accuracy while preserving the spectral and spatial details of the original image.Experimental results validate the effectiveness and robustness of the proposed method,demonstrating its superior performance compared to state-of-the-art techniques in terms of visual quality and quantitative metrics.The integration of CBAM and multi-scale attention mechanism proves instrumental in achieving these results,underscoring their significance in enhancing model performance.The proposed method offers a promising solution for improving remote sensing data quality.Future research will concentrate on optimizing the model architecture and expanding its applicability to additional types of cloud pollution and remote sensing datasets.

To address the challenges of missed detections caused by insufficient shape and texture features and blurred boundaries in existing detection methods, this paper introduces a novel moving vehicle detection approach for satellite videos. The proposed method leverages frame difference and convolution to effectively integrate spatiotemporal information. First, a frame difference module (FDM) is designed, combining frame difference and convolution. This module extracts motion features between adjacent frames using frame difference, refines them through backpropagation in the neural network, and integrates them with the current frame to compensate for the missing motion features in single-frame images. Next, the initial features are processed by a backbone network to further extract spatiotemporal feature information. The neck incorporates deformable convolution, which adaptively adjusts convolution kernel sampling positions, optimizing feature representation and enabling effective multiscale information integration. Additionally, shallow large-scale feature maps, which use smaller receptive fields to focus on small targets and reduce background interference, are fed into the detection head. To enhance small-target feature representation, a small-target self-reconstruction module (SR-TOD) is introduced between the neck and the detection head. Experiments using the Jilin-1 satellite video dataset demonstrate that the proposed method outperforms comparison models, significantly reducing missed detections caused by weak color and texture features and blurred boundaries. For the satellite-video moving vehicle detection task, this method achieves notable improvements, with an average F1-score increase of 3.9% and a per-frame processing speed enhancement of 7 s compared to the next best model, DSFNet.

Objective Optical remote-sensing images are widely used in land planning,natural-resource monitoring,disaster response,and other fields owing to their timeliness,large observation range,and clear visual characteristics.However,approximately 70％of optical remote-sensing images are occluded by clouds.Cloud occlusion complicates ground-information extraction,thus severely limiting the application of optical remote-sensing images.Therefore,cloud removal from optical remote-sensing images is necessary in the preprocessing of remote-sensing images.Compared with the conventional method of removing clouds from optical remote-sensing images,cloud removal based on deep learning presents better effect and a higher accuracy,thus mitigating the issues of conventional algorithms.Recently,denoising diffusion probabilistic models(DDPMs)have attracted much attention due to their generation capabilities beyond generative adversarial networks.DDPMs are generative models that can generate high-quality images closely reflecting the distribution of training data and have achieved the best results in terms of image generation,super-resolution,segmentation,and repair.However,they require significant computing resources to perform denoising.By contrast,the latent diffusion model can obtain high-quality images under less demanding computing requirements.Therefore,this study proposes a cloud-removal method based on the latent diffusion model to remove cloud occlusion from optical remote-sensing images and restore their surface information. Methods The cloud-removal method using the hidden diffusion model for optical remote-sensing images proposed herein is outlined as follows:first,a perceptual compression model is used to learn a hidden space on a cloudless remote-sensing image,and a hidden-space perception equivalent to the original pixel space is established.Training the DDPM in a hidden space can reduce the computing requirements and ensure high-quality image generation.Subsequently,a cloudy image is added to the hidden space to guide the diffusion model to generate a cloudless image,and noise estimation is performed using a U-Net-like cross-covariance self-attention noise estimation network(NEUTViT).The NEUTViT includes a jump connection,cross-covariance attention mechanism,and gated linear unit,which can effectively utilize low-level features,significantly reduce the computational burden,improve the nonlinear characterization ability,and achieve more accurate noise estimation.Additionally,the loss of a similar structural constraint is introduced in the forward process to alleviate the randomness of model generation and guide the model to generate cloudless images closer to the source image,thereby achieving a better cloud-removal effect. Results and Discussions First,the possibility of applying the latent diffusion model for removing clouds in optical remote-sensing images is investigated.The proposed method is evaluated on the STGAN and SEN12MS-CR Winter datasets.On the STGAN dataset,the signal-to-noise ratio and structural similarity are 26.706 and 0.759,respectively,which are 9.855 and 0.171 higher than those yielded by the comparison method on average.On the SEN12MS-CR Winter dataset,the signal-to-noise ratio and structural similarity are 28.779 and 0.798,respectively,which are 7.683 and 0.124 higher than those yielded by the comparison method on average.Experimental results show that the proposed method is superior to the comparison method and can remove clouds in optical remote-sensing images more effectively(T ables 1 and 2).The cloud-removed image yielded by this method offers three advantages:1)high color fidelity;2)favorable textural-detail preservation;3)considerable ability to remove shadows caused by clouds(Figs.6 and 7). Second,we discuss the effects of the cross-covariance attention mechanism and gated linear units on a noise-estimation network.Experiments show that the cross-covariance attention not only improves the noise estimation ability of the noise-estimation network but also significantly reduces the computational complexity of the model.The gated linear unit effectively reduces the computational complexity of the network and enhances the cloud-removal ability of the model(Table 3). Finally,a model using a single loss L2 and another model using joint loss(L2+LSSIM)were compared.The model using the joint loss achieved better results.Compared with the single loss L2,the joint loss enhances the global-structure-recovery ability of the model and improves the quality of the cloud-free image generated by the model(Table 4). Conclusions Cloud removal in remote-sensing images is mandatory in the preprocessing of remote-sensing images and has been investigated extensively.This paper proposes a method to remove clouds from optical remote-sensing images using a hidden diffusion model and restore their surface information.In the forward process,a structural-similarity constraint loss is introduced to alleviate the randomness of model generation,and a U-Net-like cross-covariance attention noise estimation network(NEUTViT)is proposed to estimate the noise distribution more accurately.The cloud removal results obtained on two datasets outperform that of other single-image remote sensing cloud removal methods.The LDMCR model proposed herein performs better than other similar methods;however,it presents some limitations.For example,it can not easily reconstruct surface information enshrouded by large,thick clouds and does not use additional data as aid.In the future,we will use auxiliary data(such as SAR images)and combine them with the cloud-removal tasks of large-scale optical remote-sensing images to investigate cloud removal from optical remote-sensing images using the latent diffusion model.

Convolutional neural networks (CNNs) extract semantic features from images by stacking convolutional operators, which easily causes semantic information loss and leads to hollow and edge inaccuracies in building extraction. Therefore, a features self-attention U-block network (FSAU-Net) is proposed. The network focuses on the target feature self-attention in the coding stage, and features self-attention (FSA) distinguishes buildings from nonbuilding by weighting the extracted features themselves; we introduce spatial attention (SA) in the decoder stage to focus on the spatial locations of features, and SA generates spatial location features through the spatial relationship among the features to highlight the building information area. A jump connection is used to fuse the shallow features generated in the decoder stage with the deep features generated in the encoder stage to reduce the building information loss. We validate the superiority of the method FSAU-Net on the WHU and Inria datasets with 0.3 m resolution and Massachusetts with 1.0 m resolution, experimentally showing IoU of 91.73%, 80.73% and 78.46% and precision of 93.60%, 90.71% and 86.37%, respectively. In addition, we also set up ablation experiments by adding an FSA module, Squeeze-and-Excitation (SE) module and Efficient Channel Attention (ECA) module to UNet and ResNet101, where UNet+FSA improves the IoU values by 3.15%, 2.72% and 1.77% compared to UNet, UNet+SE and UNet+ECA, respectively, and ResNet101+FSA improves the IoU values by 2.06%, 1.17% and 0.9% compared to ResNet101, ResNet101+SE and ResNet101+ECA, respectively, demonstrating the superiority of our proposed FSA module. FSAU-Net improves the IoU values by 3.18%, 2.75% and 1.80% compared to those of UNet, UNet+SE and UNet+ECA, respectively. FSAU-Net has 2.11%, 1.22%, and 0.95% IoU improvements over the IoU values of ResNet101, ResNet101+SE and ResNet101+ECA, respectively, demonstrating the superiority of our proposed FSAU-Net model. The TensorFlow implementation is available at .

积雪对气候高度敏感,能够对气候变化进行响应,对积雪变化的研究具有重要意义.本文基于 Landsat 7 ETM SLC、Landsat 8 OLI影像,利用eCongnition面向对象提取与人机交互解译方法提取并分析 2000 和 2020 年云南地区积雪分布数据,然后结合 23 个气象站点、50 个插值点在 2000-2020 年的温度、降水、降雪量和雪深记录、高程数据和等辅助数据,使用一元线性回归分析法、Mann-Kendall趋势检验、突变检测法,进行云南地区积雪分布现状、2000-2020 年积雪变化情况及气候响应分析.结果表明:①2000 年云南地区积雪总面积为 1 533.1 km2,2020 年为 987.8 km2,2000-2020 年,积雪面积总体减少 545.3 km2.其中积雪减少部分面积为 945.6 km2,积雪增加部分面积为400.3002 km2,保持不变的为587.4633 km2.②云南地区积雪分布与变化表现出显著的空间差异性.积雪面积变化集中分布在28°N—28.5°N和3.6～4.0km高程范围内,面积分别减少16.2%和 14.5%.③云南地区的温度和降水均呈显著上升趋势,每10 年线性倾向量分别达0.38℃和314.7 mm,与积雪面积变化呈负相关;降雪量和雪深呈弱下降趋势,与积雪总体变化呈正相关.该研究为云南地区积雪空间分布与变化、气候变化趋势提供了支撑数据.

针对深层卷积较难兼顾全局特征与局部特征从而导致提取建筑物边缘不准确和微小建筑物丢失的问题,以注意力机制和跳跃连接为基础提出SER-UNet算法.SER-UNet算法在编码器阶段耦合SE-ResNet和最大池化层,在解码器阶段关联SE-ResNet与反卷积层,通过跳跃连接将编码器提取的浅层特征和解码器提取的深层特征进行融合后输出特征图.验证SER-UNet算法的有效性,在MAP-Net网络并行多路径特征提取阶段使用SER-UNet算法替换原网络中的特征提取结构,分别在WHU数据集和Inria数据集上进行评估,IoU与精度分别达91.46％、82.61％和95.67％、92.75％,对比UNet、PSPNet、ResNet101,MAP-Net 网络,IoU 分别提高 0.49％、0.14％、1.89％、1.57％,精度分别提高0.14％、1.06％、2.42％、1.09％.分析 SER-UNet算法的泛化能力,将级联 SER-UNet 的 MAP-Net 网络在Aeriallmage数据集上进行提取验证,IoU与精度分别达85.32％和94.13％.结果表明,结合SER-UNet算法的MAP-Net并行多路径网络表现出较好的提取精度与泛化能力.此外,SER-UNet算法可以有效地嵌入PSPNet、ResNet101、HRNetv2等网络中,提升网络特征表示能力.

针对连续池化操作丢失低层语义信息而导致建筑物提取精度低的问题,尝试以UNet++网络为基础,通过将编码器的标准卷积及最大池化替换成深度可分离卷积,以及在编码器末端利用不同采样率的空洞卷积构建多尺度空洞空间金字塔池化结构ASPP(Atrous Spatial Pyramid Pooling)来提升网络性能,并将改进后的建筑物提取网络称为残差空洞空间金字塔网络(Res_ASPP_UNet++).为验证Res_ASPP_UNet++网络结构的有效性和适用性,以经过数据增强预处理的WHU和Massachusetts数据集作为数据源,对Res_ASPP_UNet++网络与目前常用的语义分割网络进行了试验和精度评估,并将Res_ASPP_UNet++网络与文献中的研究成果进行了对比.结果表明Res_ASPP_UNet++在模型参数量与精度两个方面均表现出优势,能够在大幅压缩模型参数量的前提下,显著提升建筑物提取精度,提取建筑物的边界更加平滑和精确,对不同尺度的建筑物表现出较强的泛化能力.

In order to accurately reconstruct the NDVI time series polluted by continuous clouds and fog and to improve the fidelity of the fitted curve, this paper proposes a weighted SG filtering method based on the Dixon test and the fusion of spatiotemporal information. Firstly, use the Dixon test to generate a standard reference curve for the pixel. Secondly, the reference curve and spatiotemporal information are used to find similar pixels near the target pixel, and the preprocessing sequence is generated by integrating similar pixel data. Finally, the processed NDVI sequence is smoothed by the weighted SG filtering method to obtain high-quality NDVI time series data. MODS13Q1＿NDVI data from 2010-2019 were used for comparison with the SG filter method, the double logistic fitting method, and the asymmetric Gaussian function method. The results show that when fitting the curve, the SGDST method can accurately reconstruct the time series of missing data for multiple consecutive periods, and effectively retain the details of local peaks and troughs; when there is noise, it can avoid over-correcting the effective low value. Compared with the other three methods, the SGDST method has the largest proportion of root mean square error and mean absolute error which are less than 0.085 3 pixels in the total fitted pixels and the best fidelity.

针对光学遥感影像受天气影响导致部分地物间形成弱边缘问题,提出一种低维纹理特征算子与双变异蝴蝶优化算法.首先提出一种适用于遥感影像的低维完备局部三值模式的纹理特征提取算子,并将其引入简单线性迭代聚类算法,对遥感影像进行初始分割,减小了噪音影响,同时增强算法对弱边缘的敏感度和分割准确性;然后采用双变异蝴蝶优化的支持向量机合并同质超像素块,以简单线性迭代聚类算法和低维纹理特征算子得到的综合特征作为输入,得到最终分割图像.利用2组高分辨率遥感影像进行分割实验,并与当下流行的卷积神经网络进行对比,实验结果表明,所提算法相较于传统算法对弱边缘有更好的分割效果,数据一的边界回归率(boundary recall,BR)值较对比算法平均提高了 1.9％,Kappa系数平均提高了 0.036;数据二的BR值较对比算法平均提高了 2.33％,Kappa系数平均提高了 0.027.对比实验证明了所提算法相较于卷积神经网络有更好的泛化性.

针对相机大视场感兴趣目标分辨率不足的问题,提出短基线同轴约束模型并设计主从相机原型,大视场相机用于监视整个视场,主动相机用于对目标区域进行指向性高清观测.基本过程为:(1)利用短基线同轴约束模型简化相机外参矩阵,并构建大视场相机与主动相机间的映射关系,通过三角函数计算主动相机初始控制参数;(2)在近距离场景下,对主动相机控制参数进行补偿.实验结果表明,相机只需一次离线标定即可适应各种场景.观测目标与大视场相机距离3～10 m的近距离场景下,目标在主动相机图像中的实际位置与理论位置误差在30像素以内;远距离场景的有效距离内,误差在6像素左右.计算单一目标点对应主动相机控制参数的时间不超过0.2 ms.原型对目标场景、目标深度无依赖性,且用于较远目标观测时相对于其他方法在精度与时效性方面具有较高优势.

使用人工模拟的高-低分辨率图像对易导致在对真实遥感图像超分辨率重建时模型泛化能力差,针对此问题,结合残差通道注意力网络(residual channel attention network,RCAN)的二次残差(residual in residual,RIR)模块,改进对偶回归网络(dual regression networks,DRN),提出了残差对偶回归网络(residual dual regression network,RDRN).选取LandCover.ai和DIOR航空图像数据集的10000张512像素×512像素图像构成样本数据集,用于训练和测试网络,并将重建结果与现有其他超分辨率网络模型的重建结果对比评价.实验结果表明,RDRN在重建质量和模型参数量方面均表现优异,能够在较低模型复杂度的情况下实现较好的超分重建效果,且对不同低分辨率遥感图像具有较好的泛化能力.

针对利用土地覆盖与数字高程模型生成遥感影像存在部分细节缺失的问题,该文提出结合空洞卷积的条件对抗网络以生成高分辨率遥感影像.首先,生成器网络采用空间 自适应归一化结构将土地覆盖特征与数字高程模型特征融合;其次,将并行空洞卷积模块嵌入生成器网络中的编码器中,以改善融合后的特征;最后,解码器输出的影像输入判别器网络进行判别,输出生成结果.为验证该文网络的有效性,与现有的条件生成对抗网络在GeoNRW和DFC2020数据集上进行实验.结果表明:该文网络在GeoNRW数据集上的距离得分较对比网络平均降低0.11,整体平均交并比较对比网络平均提升0.77％,生成的影像地物细节更丰富;此外,该文网络在DFC2020数据集上具有一定的泛化能力,生成的影像优于对比网络生成的影像.

在运动控制应用场景中,由于疲劳等原因会导致设备微小形变,因此需要更加密集的观测手段与恢复方法.考虑到待检设备的摄影测量属性及高帧频相机高信息量特性,利用240帧/s的高帧频电荷耦合元件(charge coupled device,CCD)相机结合摄影测量方法实现对运动控制轴形变信息的精确量测,具体步骤如下:(1)固定标定板,由运动控制器控制高帧频相机运动,并对标定板进行连续拍摄.(2)提取标定板角点坐标,提出基于角点几何与运动特征的角点坐标精确定位方法.(3)计算相机坐标序列,恢复相机运动的高密度点轨迹.实验结果表明,由轨迹可精确地反映室内运动控制轴形变位置和形变量,实现亚毫米级形变检测,实际检测的运动轴最小倾斜角为0.012°,水平轴左端48.016 cm处存在0.98 mm的形变,竖直轴上端3.810 cm处形变量为2.00 mm.

卷烟烟包外包装图案缺失检测是卷烟生产中的重要环节,目前常用的图案缺失检测方法普遍存在漏检率高、识别精度不够和速度慢等问题.因此本文提出了一种基于YOLOv4-tiny的烟包缺陷快速检测方法.该方法主要包括:(1)使用CSPDarknet53-tiny对图像特征进行提取.(2)通过添加多空间空洞卷积融合模块,获得丰富的上下文信息,增强感受野,从而提高模型的检测精度.(3)使用EIoU边框位置回归损失函数,提高烟盒图案识别预测框的精确率.本文方法与YOLOv4-tiny进行对比,实验结果表明,本文算法mAP值为97.35％,检测烟包外观的平均时间为17 ms,能够满足卷烟小包外观检测对时间和精确率的要求.相较于YOLOv4-tiny在mAP上提升了1.34％,在Average IoU上提升了3.68％,速度基本与YOLOv4-tiny持平,在保持快速检测的同时能够有效的提高精度.

Associated traveling becomes new research highlight in the fields of urban planning, traffic traveling, infectious disease prevention and controlling, crime investigation etc. Especially, effectively identifying the related behavior of individuals with subjective intentions (such as meeting and waiting) is considered as a difficult problem in the area of spatio-temporal cognition. In this paper, spatio-temporal trajectory of crime is studied. First, the inadequacy of the previous traveling trajectory description methods is analyzed, which indicating that the detailed point characteristics and the complete line characteristics of the traveling trajectory should be considered, based on that a travel trajectory description model considering the point-line characteristics is proposed. Second, a trajectory discrimination method of associated traveling is proposed based on spatio-temporal topological relationship, which can describe four basic associated traveling trajectory modes including meeting, waiting, coexistence and companion, further distinguishing 19 different associated traveling subtypes. Finally, the effectiveness of the method is validated by an experimental comparison.

针对如何以端到端可训练的方式重构建筑物轮廓的问题,本文提出了一种端到端多残差模块堆叠沙漏网络的建筑物轮廓重构方法.首先,采用多残差模块堆叠沙漏网络提取建筑物角点和边缘特征;其次,利用角点检测模块匹配对应角点的相对位置,以获取候选角点;然后,通过线段采样将候选角点生成候选轮廓线;最后,线验证模块利用候选线段及多残差模块堆叠沙漏网络得到特征图,并验证每个线段是否为建筑物轮廓线,以获得建筑物轮廓重构结果.试验结果表明,在SpaceNet建筑物数据集上,本文方法能检测出建筑物角点及边缘,并有效实现了以端到端可训练的方式重构建筑物轮廓.

In semi-supervised semantic segmentation, the segmentation accuracy of aerial images is mainly improved by using the structure of encoder—master-auxiliary decoder which applies the unlabeled samples to the calculation. However, the loss of shallow detail features which is caused by continuous downsampling in the process of encoding makes the boundary of ground objects incomplete. Therefore, a semi-supervised network combining multi-scale shared encoding is proposed for semantic segmentation of aerial images. The encoder uses ResNet-50 to obtain the shallow features of the image, and links the shallow features by embedding a multi-scale shared coding module at the end of ResNet-50 to build a dense feature pyramid and expand the receptive field, thereby obtaining multi-scale detailed information of the target feature. The effectiveness of the proposed method is verified by compared with UNet, DeepLabv3+, FCN and CCT, XModalNet, VLCNet on the two datasets of LandCover.ai and DroneDeploy, and the result shows that our network has obvious advantages in terms of label number and accuracy. For the LandCover.ai dataset, under the premise of 6000 labeled samples and 6500 unlabeled samples, the overall mIoU increased by 1.15%. For the DroneDeploy dataset, under the premise of 30 labeled samples and 5 unlabeled samples, the overall mIoU increased by 0.94%, while significantly improving the segmentation accuracy of ground objects to obtain a clear and complete ground boundary.

模糊航空图像复原不仅能够改善图像的细节与特征,而且可以提高目标的识别能力以及定位精度.在无参数条件下利用已有航空图像建立模糊核估计模型,提出了顾及模糊核连通性的无人机图像半盲复原方法.首先建立梯度筛选,筛选出模糊图像与已有清晰图像梯度域的公共地物,构建保真项;然后利用模糊核梯度八邻域描述模糊核的连通性度量,并将其作为正则项缩小解空间与构建模型;最后重建图像,根据图像金字塔结构对模糊核进行分层估计,并通过分裂Bregman算法解卷积重建图像.对比实验从模糊类型、公共地物、方法对比、方法适用性4个方面进行分析,结果表明,在公共地物达到35％以上时,模糊航片具有良好的复原效果,所提方法具有较强实际应用价值.