Change detection represents a fundamental research area within remotesensing technology. Nevertheless, in practical applications, the spatial arrangement of terrestrial features is frequently highly complex. The chan...
详细信息
Land use classification using optical and Synthetic Aperture Radar (SAR) images is a crucial task in remotesensingimage interpretation. Recently, deep multi-modal fusion models have significantly enhanced land use c...
详细信息
ISBN:
(纸本)9789819985487;9789819985494
Land use classification using optical and Synthetic Aperture Radar (SAR) images is a crucial task in remotesensingimage interpretation. Recently, deep multi-modal fusion models have significantly enhanced land use classification by integrating multi-source data. However, existing approaches solely rely on simple fusion methods to leverage the complementary information from each modality, disregarding the intermodal correlation during the feature extraction process, which leads to inadequate integration of the complementary information. In this paper, we propose FASONet, a novel multi-modal fusion network consisting of two key modules that tackle this challenge from different perspectives. Firstly, the feature alignment module (FAM) facilitates cross-modal learning by aligning high-level features from both modalities, thereby enhancing the feature representation for each modality. Secondly, we introduce the multi-modal squeeze and excitation fusion module (MSEM) to adaptively fuse discriminative features by weighting each modality and removing irrelevant parts. Our experimental results on the WHU-OPT-SAR dataset demonstrate the superiority of FASONet over other fusion-based methods, exhibiting a remarkable 5.1% improvement in MIoU compared to the state-of-the-art MCANet method.
Simultaneous localization and mapping (SLAM) can provide point clouds and self-position data for 3D mapping, virtual reality, augmented reality, mixed reality, UAV flight control, and autonomous vehicle control. In 3D...
详细信息
Simultaneous localization and mapping (SLAM) can provide point clouds and self-position data for 3D mapping, virtual reality, augmented reality, mixed reality, UAV flight control, and autonomous vehicle control. In 3D mapping, 3D object recognition based on model fitting and machine learning is required for the automation of 3D mapping and modeling using dense point clouds generated by SLAM and mobile mapping. The machine learning-based 3D object recognition is classified into clustering and segmentation. The segmentation is fundamental processing for 3D mapping, scan-to-BIM, and autonomous vehicles using point clouds. The segmentation can be classified into image-based and 3D-based approaches. Each approach has strengths and weaknesses, thus we focus on the integration of image-based and 3D-based approaches embedded in SLAM and mobile mapping. In this research, we propose a methodology to improve the performance of streaming point cloud processing based on image-based and 3D-based point cloud segmentation for 3D mapping of urban river environments. We also developed a methodology of streaming point cloud segmentation embedded in GNSS/LiDAR-SLAM and multi-beam scanning. In our experiments, we used a water-borne mobile mapping system at an urban river as GNSS and non-GNSS environments. We acquired dense streaming point clouds above water surfaces with GNSS/LiDAR-SLAM consisting of two LiDARs and precise point positioning based on real-time kinematic positioning with a centimeter-level augmentation service using a quasi-zenith satellite system. In parallel, we also acquired dense streaming point clouds underwater surfaces with a multi-beam scanning sonar with RTK-GNSS positioning. Moreover, we experimented with streaming point cloud segmentation of acquired massive point clouds to classify streaming point clouds into bridges, revetments, buildings, and underwater surfaces. Through the experiment using the streaming point clouds, we confirmed that our methodology can i
Deep Learning classification of nine imagery typologies (categories) of desert-fringe natural vegetation is performed using DenseNet with six input layer types: RGB, Red, Red CDF (Cumulative Distribution Function), Re...
详细信息
ISBN:
(纸本)9798350345421
Deep Learning classification of nine imagery typologies (categories) of desert-fringe natural vegetation is performed using DenseNet with six input layer types: RGB, Red, Red CDF (Cumulative Distribution Function), Red Edge (variance), Combined (Red, CDF & Edge), and VARI (a measure of greenness). F1-score accuracy higher than 0.95 within 50 epochs was found for the RGB imagery. Red band and CDF alone showed only slightly lower performance. VARI representing 'net' color information facilitated very low separability between the nine pattern categories. However, combining Red, Red CDF and Red Edge allowed better classification than RGB, and a more stable increase in F1-scores was seen with the increase in the number of epochs. These results suggest that simple image morphological pre-processing may improve deep learning classification performance. Yet, these results are very preliminary and encourage the use of other mathematical morphology algorithms.
Most of the land surface water in the plateau region is frozen. Due to the huge difference in the dielectric constant of water and ice, this paper generated a ten-year surface permafrost change map from 1992 to 2001 b...
详细信息
High-resolution remotesensingimage segmentation is a mature application in many industrial-level image applications, such as those in the military, civil, and other fields. Scene analysis needs to be automated in hi...
详细信息
ISBN:
(纸本)9789811657344
High-resolution remotesensingimage segmentation is a mature application in many industrial-level image applications, such as those in the military, civil, and other fields. Scene analysis needs to be automated in high-resolution remotesensingimages as much as possible. Nowadays, with the rise of deep learning algorithms, remotesensingimageprocessing algorithms have made tremendous progress. Deep learning algorithms process unlabeled data by learning a certain amount of labeled data. We conducted a specific study on the road target with GF1 data collected in China, and the remotesensingimage’s resolution was 2 m (Jin et al. in J Anhui Agri Sci 43:358–362, 2015 [Jin et al. in J Anhui Agri Sci 43:358–362, 2015]). According to the observation of road features in remotesensingimages, it still has a large number of small roads that are difficult to distinguish in the 2 m resolution GF1 remotesensingimage. Due to the limitations of the downsampling calculation of the fully convolutional neural network, it is easy to lose a lot of information on small roads (Long et al. in Proceedings of the IEEE conference on computer vision and patternrecognition, pp. 3431–3440, 2015 [Long J, Shelhamer E, Darrell T (2015) Fully convolutional networks for semantic segmentation. In: Proceedings of the IEEE conference on computer vision and patternrecognition, 2015, pp 3431–3440]). Therefore, we have adjusted the feature extraction and backbone networ. We adopted EfficientNet (Tan and Le in Efficientnet: Rethinking model scaling for convolutional neural networks, 2019 [Tan M, Le QV (2019) Efficientnet: rethinking model scaling for convolutional neural networks. arXiv preprint arXiv:1905.11946]) as the skeleton network of the algorithm, and combined D-linkNet (Zhou et al. in CVPR workshops, pp. 182–186, 2018 [Zhou L, Zhang C, Wu M (2018) D-LinkNet: LinkNet with pretrained encoder and dilated convolution for high resolution satellite imagery road extraction. In: CVPR workshops,
In remotesensingimage classification, it is difficult to distinguish the homogeneity of same land class and the heterogeneity between different land classes. Moreover, high spatial resolution remotesensingimages o...
详细信息
ISBN:
(纸本)9783031189128;9783031189135
In remotesensingimage classification, it is difficult to distinguish the homogeneity of same land class and the heterogeneity between different land classes. Moreover, high spatial resolution remotesensingimages often show the phenomenon of ground object classes fragmentation and salt-and-pepper noise after classification. To improve the above phenomenon, Markov random field (MRF) is a widely used method for remotesensingimage classification due to its effective spatial context description. Some MRF-based methods capture more image information by building interaction between pixel granularity and object granularity. Some other MRF-based methods construct representations at different semantic layers on the image to extract the spatial relationship of objects. This paper proposes a new MRF-based method that combines multi-granularity and different semantic layers of information to improve remotesensingimage classification. A hierarchical interaction algorithm is proposed that iteratively updates information between different granularity and semantic layers to generate results. The experimental results demonstrate that: on the Gaofen-2 imagery, the proposed model shows a better classification performance than other methods.
remotesensing technologies allow for continuous and valuable monitoring of the Earth’s various environments. In particular, coastal and ocean monitoring presents an intrinsic complexity that makes such monitoring th...
详细信息
Capturing motion vehicle information from satellite videos is crucial for real-time traffic monitoring and emergency response. However, vehicles in satellite videos are small in size, lack detailed textural features a...
详细信息
Conventional imaging detection relies on electrical modulation, but it grapples with limitations in speed and adaptability within complex environments. Intelligent imaging detection, capable of manipulating micro-nano...
详细信息
暂无评论