Within (semi-)automated visual industrial inspection, learning-based approaches for assessing visual defects, including deep neural networks, enable the processing of otherwise small defect patterns in pixel size on h...
详细信息
ISBN:
(纸本)9798350318920;9798350318937
Within (semi-)automated visual industrial inspection, learning-based approaches for assessing visual defects, including deep neural networks, enable the processing of otherwise small defect patterns in pixel size on high-resolution imagery. The emergence of these often rarely occurring defect patterns explains the general need for labeled data corpora. To alleviate this issue and advance the current state of the art in unsupervised visual inspection, this work proposes a DifferNet-based solution enhanced with attention modules: AttentDifferNet. It improves image-level detection and classification capabilities on three visual anomaly detection datasets for industrial inspection: InsPLAD-fault, MVTec AD, and Semiconductor Wafer. In comparison to the state of the art, AttentDifferNet achieves improved results, which are, in turn, highlighted throughout our quali-quantitative study. Our quantitative evaluation shows an average improvement compared to DifferNet - of 1.77 +/- 0.25 percentage points in overall AUROC considering all three datasets, reaching SOTA results in InsPLAD-fault, an industrial inspection in-the-wild dataset. As our variants to AttentDifferNet show great prospects in the context of currently investigated approaches, a baseline is formulated, emphasizing the importance of attention for industrial anomaly detection both in the wild and in controlled environments.
Due to the complexity of the composition of surface features information in remotesensingimages, the recognition and mapping effects are often difficult to meet the actual application needs. Therefore, the research ...
详细信息
ISBN:
(数字)9798350355413
ISBN:
(纸本)9798350355420
Due to the complexity of the composition of surface features information in remotesensingimages, the recognition and mapping effects are often difficult to meet the actual application needs. Therefore, the research on remotesensingimage surface features recognition and vectorization mapping based on artificial intelligence is proposed. In the construction stage of remotesensingimage feature map, based on the internal relationship between remotesensing data points, the feature map is used to reflect the local similarity characteristics of remotesensing data; Through the semi supervised mechanism of artificial intelligence algorithm - Markov random field model, the specific information of the ground object map is recognized. In the mapping stage, the constrained geometry algorithm is introduced, and the CGA rule file is used as a component to drive the 3D creation process. After the integration of basic terrain and 2D spatial data, the mapping is completed by aligning the footprint of the figure recognition object with the terrain network. In the test results, the OA corresponding to the recognition results of the design method has obvious advantages in stability and specific level compared with the control group, and mapping can effectively reflect the details of ground objects.
This research is mainly concerned with enhancing the object recognition and classification of multispectral imagery imagery especially of people and vehicles in the heavily vegetated regions. Sub-technology such as PC...
详细信息
In the past decade, various haze removal techniques have been widely reported for object recognition. But hitherto little has been identified on the use of single image dehazing using transfer learning approach for ob...
详细信息
The presence of noise, displacement of points, and empty spots in a raw Light Detection and Ranging (LiDAR) point cloud are common phenomena caused by reflective surfaces or objects. Typical approaches to solve this p...
详细信息
ISBN:
(纸本)9781510655386;9781510655379
The presence of noise, displacement of points, and empty spots in a raw Light Detection and Ranging (LiDAR) point cloud are common phenomena caused by reflective surfaces or objects. Typical approaches to solve this problem are either avoid or cover the reflective areas or to manually remove the erroneous data in post processing. This can help clean the point cloud structure but will cause sparsity issues. To combat this, in this paper, we introduce a two-step process to perform point cloud restoration. Instead of removing noise, this approach can restore the points to the closest surface which they may belong to. Next, to fill out empty spots, we introduce a technique called point cloud inpainting, which involves interpolating points in 2D then mapping it back to 3D for flat surfaces. The point cloud then becomes more photorealistic and easier to use for other computer vision tasks.
Application of hyperspectral remotesensing technology in geological identification has highlighted its advantages, but the identification accuracy of metal mineral cations is limited. In this paper, the dominant spec...
详细信息
The deep neural network (DNN) has made significant progress in the single remotesensingimage super-resolution (SRSISR). The success of DNN-based SRSISR methods mainly stems from the use of the global information and...
详细信息
ISBN:
(纸本)9783031189159;9783031189166
The deep neural network (DNN) has made significant progress in the single remotesensingimage super-resolution (SRSISR). The success of DNN-based SRSISR methods mainly stems from the use of the global information and the fusion of shallow features and the deep features, which fits the non-local self-similarity characteristic of the remotesensingimage very well. However, for the fusion of different depth (level) features, most DNN-based SRSISR methods always use the simple skip-connection, e.g. the element-wise addition or concatenation, to transform the feature coming from preceding layers to later layers directly. To achieve sufficient complementation between different levels and capture more informative features, in this paper, we propose a stage-mutual-affine network (SMAN) for high-quality SRSISR. First, for the use of the global information, we construct a convolution-transformer dual-branch module (CTDM), in which we propose an adaptive multi-head attention (AMHA) strategy to dynamically rescale the head-wise features of the transformer for more effective global information extraction. Then, the global information is fused with the local structure information extracted by the convolution branch for more accurate recurrence information reconstruction. Second, a novel hierarchical feature aggregation module (HFAM) is proposed to effectively fuse shallow features and deep features by using a mutual affine convolution operation. The superiority of the proposed HFAM is that it achieves sufficient complementation and enhances the representational capacity of the network by extracting the global information and exploiting the interdependencies between different levels of features, effectively. Extensive experiments demonstrate the superior performance of our SMAN over the state-of-the-art methods in terms of both qualitative evaluation and quantitative metrics.
Aerial scene images are often imbalanced, where the most common classes as majorities and a few significant classes as minorities. We observe that the majority classes not only dominate the classification optimization...
详细信息
ISBN:
(纸本)9783031301100;9783031301117
Aerial scene images are often imbalanced, where the most common classes as majorities and a few significant classes as minorities. We observe that the majority classes not only dominate the classification optimization but also generate deviations that affect the classifier weight matrices. In this work, we propose a hybrid framework based on classifier calibration, which mitigate the effect of the class imbalance problem in aerial scene recognition. In particular, the framework progressively incorporates feature representation and classifier learning branches, while building a memory bank of learned representations for approximating deviations derived from imbalanced data. We calibrate the classifier by excluding the deviations in the prediction of the testing stage. Extensive experiments are evaluated on class imbalanced aerial scene image datasets, which show the advantages of the proposed hybrid framework with classifier calibration outperforming state-of-the-art aerial scene recognition methods.
Crop area mapping as well as other remotesensing - based agricultural applications are vital for food security and production. This review investigates several studies published in two last years which focus on crop ...
详细信息
Change detection is one of the main applications of remotesensingimages. Pixel-to-pixel change detection using deep learning has been a hot research spot. However, the current approach are not effective enough to fu...
详细信息
ISBN:
(纸本)9783031189159;9783031189166
Change detection is one of the main applications of remotesensingimages. Pixel-to-pixel change detection using deep learning has been a hot research spot. However, the current approach are not effective enough to fuse deep semantic features and raw spatial information, and the network does not have the ability to perform long-distance information aggregation due to the limitation of the convolutional kernel size. In this manuscript, we propose a Siamese UNet with a dense attention mechanism, named SUDANet to do change detection for remotesensingimages. SUDANet add a channel attention mechanism and a self-attention mechanism to the dense skip connection between encoder and decoder which enable the model to fuse feature information in channel dimensions and spatial dimensions. Graph attention module is also added at the end of the encoder, enabling the model to perform correlation analysis and long-distance aggregation of deep semantic features. The experimental results on LEVIR dataset show that our method outperforms the state-of-the-art change detection methods.
暂无评论