Remote sensing image segmentation with samples imbalance is always one of the most important issues. Typically, a high-resolution remote sensing image has the characteristics of high spatial resolution and low spectral resolution, complex large-scale land covers, small class differences for some land covers, vague foreground, and imbalanced distribution of samples. However, traditional machine learning algorithms have limitations in deep image feature extraction and dealing with sample imbalance issue. In the paper, we proposed an improved full-convolution neural network, called DeepLab V3+, with loss function based solution of samples imbalance. In addition, we select Sentinel-2 remote sensing images covering the Yuli County, Bayingolin Mongol Autonomous Prefecture, Xinjiang Uygur Autonomous Region, China as data sources, then a typical region image dataset is built by data augmentation. The experimental results show that the improved DeepLab V3+ model can not only utilize the spectral information of high-resolution remote sensing images, but also consider its rich spatial information. The classification accuracy of the proposed method on the test dataset reaches 97.97%. The mean Intersection-over-Union reaches 87.74%, and the Kappa coefficient 0.9587. The work provides methodological guidance to sample imbalance correction, and the established data resource can be a reference to further study in the future.
Remote image semantic segmentation technology is one of the core research elements in the field of computer vision and has a wide range of applications in production life. Most remote image semantic segmentation methods are based on CNN. Recently, Transformer provided a view of long-distance dependencies in images. In this paper, we propose RCCT-ASPPNet, which includes the dual-encoder structure of Residual Multiscale Channel Cross-Fusion with Transformer (RCCT) and Atrous Spatial Pyramid Pooling (ASPP). RCCT uses Transformer to cross fuse global multiscale semantic information; the residual structure is then used to connect the inputs and outputs. ASPP based on CNN extracts contextual information of high-level semantics from different perspectives and uses Convolutional Block Attention Module (CBAM) to extract spatial and channel information, which will further improve the model segmentation ability. The experimental results show that the mIoU of our method is 94.14% and 61.30% on the datasets Farmland and AeroScapes, respectively, and that the mPA is 97.12% and 84.36%, respectively, both outperforming DeepLabV3+ and UCTransNet.
Algorithms combining CNN (Convolutional Neural Network) and super-pixel based smoothing have been proposed in recent years for Synthetic Aperture Radar (SAR) image classification. However, the smoothing may lead to the damage of details. To solve this problem the feature fusion strategy is utilized, and a novel adaptive fusion module named Gated Channel Attention (GCA) is designed in this paper. In this module, the relevance between channels is embedded into the conventional gated attention module to emphasize the variation in contribution on classification results between channels of feature-maps, which is not well considered by the conventional gated attention module. A GCA-CNN network is then constructed for SAR image classification. In this network, feature-maps corresponding to the original image and the smoothed image are extracted, respectively, by feature-extraction layers and adaptively fused. The fused features are used to obtain the results. Classification can be performed by the GCA-CNN in an end-to-end way. By the adaptive feature fusion in GCA-CNN, the smoothing of misclassification and the detail keeping can be realized at the same time. Experiments have been performed on one elaborately designed synthetic image and three real world SAR images. The superiority of the GCA-CNN is demonstrated by comparing with the conventional algorithms and the relative state-of-the-art algorithms.
Natural reserves play a leading role in safeguarding national ecological security. Remote sensing change detection (CD) technology can identify the dynamic changes of land use and warn of ecological risks in natural reserves in a timely manner, which can provide technical support for the management of natural reserves. We propose a CD method (CA-BIT) based on the improved bitemporal image transformer (BIT) model to realize the change detection of remote sensing data of Anhui Natural Reserves in 2018 and 2021. Resnet34-CA is constructed through the combination of Resnet34 and a coordinate attention mechanism to effectively extract high-level semantic features. The BIT module is also used to efficiently enhance the original semantic features. Compared with the overall accuracy of the existing deep learning-based CD methods, that of CA-BIT is 98.34% on the natural protected area CD datasets and 99.05% on LEVIR_CD. Our method can effectively satisfy the need of CD of different land categories such as construction land, farmland, and forest land.
This paper proposes convolutional neural network jointed with long short-time memory (CNN_LSTM) and Seq2Seq based on convolutional operation (Convolutional Seq2Seq), which the fully connected operation of Seq2Seq is replaced by convolution, and the attention mechanism of Seq2Seq is improved to monitor changes in water bodies. Convolutional Seq2Seq and CNN_LSTM can extract the temporal and spatial characteristics of remote sensing image time series. We also propose downsampling and resolution recovery (DDR) modules to reduce the computational resource consumption of the two models. Compared with the popular full convolutional network (FCN) −8s, DeepLab v2 with a baseline of ResNet101, and long short time memory (LSTM) methods, the water change monitoring results based on Convolutional Seq2Seq and CNN_LSTM have lower noise and higher accuracy. The CNN_LSTM method also allows fewer hidden layer features of LSTM with high-precision change monitoring results.
16 December 2024 in Artificial Intelligence for Engineering Design, Analysis and Manufacturing
Bridge engineering design drawings basic elements contain a large amount of important information such as structural dimensions and material indexes. Basic element detection is seen as the basis for digitizing drawings. Aiming at the problem of low detection accuracy of existing drawing basic elements, an improved basic elements detection algorithm for bridge engineering design drawings based on YOLOv5 is proposed. Firstly, coordinate attention is introduced into the feature extraction network to enhance the feature extraction capability of the algorithm and alleviate the problem of difficult recognition of texture features inside grayscale images. Then, targeting objectives across different scales, the standard 3 × 3 convolution in the feature pyramid network is replaced with switchable atrous convolution, and the atrous rate is adaptively selected for convolution computation to expand the sensory field. Finally, experiments are conducted on the bridge engineering design drawings basic elements detection dataset, and the experimental results show that when the Intersection over Union is 0.5, the proposed algorithm achieves a mean average precision of 93.6%, which is 3.4% higher compared to the original YOLOv5 algorithm, and it can satisfy the accuracy requirement of bridge engineering design drawings basic elements detection.