Search

Effective distress detection and quantitative analysis play a crucial role in road maintenance and driving safety. The Pavement distress segmentation network (PDSNet) is designed to combine the pyramid scene parsing network (PSPNet) and U-Net, providing both prior global information and local features that can overcome the common detection issues on the pavement data set faced by a single network. This paper proposes an efficient and improved architecture of PDSNet called PDSNet II for enhanced global modeling and retrieving fine details capacities. The proposed PDSNet II represents two major modifications on the original PDSNet. Firstly, a shifted window based on fully connected conditional random fields (FC-CRFs) layer is purposefully introduced to provide connections among consecutive self-attention layers that significantly enhance modeling power. Secondly, PDSNet II adopts multiple-head attention mechanisms to capture diverse interaction information across multiple projection spaces. Consequently, the output maps from the pyramid pooling module (PPM) head and the U-Net tail are fed into a neural window FC-CRFs layer. PDSNet II was trained using a data set consisting of 12,648 two-dimensional (2D) intensity and three-dimensional (3D) range images depicting various pavement conditions. The experimental results demonstrate that PDSNet II outperforms the original PDSNet in terms of F1-score and intersection over union (IoU). Compared with state-of-the-art networks, PDSNet II exhibits superior performance in detecting complex distress patterns, while effectively reducing noise and maintaining robustness. Overall, the proposed PDSNet II framework shows promising results in pavement distress segmentation, highlighting its potential for practical applications. [ABSTRACT FROM AUTHOR]

Accurate traffic volume prediction is crucial for intelligent transportation systems to control traffic conditions and improve travel efficiency. Traditional traffic volume prediction models focus on the similarity of passenger volume patterns in historical data. However, they ignore the continuous and periodic features of traffic volume data and the deviation in traffic volume caused by external factors such as holidays and weather. This paper proposes a multifeature fusion convolutional long-short-term memory (CLSTM) model. The model is based on a convolutional neural network (CNN) and a long-short-term memory (LSTM) neural network. The CLSTM model considers time continuity as a short-term feature, daily periodicity as a long-term feature, spatial correlation between roads as a spatial feature, and environmental factors as external features. The CNN model is applied to represent the temporal and spatial features as a two-dimensional spatial-temporal matrix, and two sets of high-level features are proposed. The fully connected neural network model is used to fuse the predictions from the feature matrix and LSTM neural networks. The effectiveness of feature extraction, model design, and model sensitivity are tested using the London M25 motorway as the research object. The results illustrate that the CLSTM model enhances both prediction accuracy and model adaptability, achieving a balance between prediction efficiency and accuracy. [ABSTRACT FROM AUTHOR]

The safety and ride comfort of the emergency rescue vehicle can be improved greatly when the prior knowledge of the terrain in front of the vehicle is used as the active suspension input. In this paper, we present a robust and precise terrain preview detection and classification system based on vision and laser fusion with deep learning. First, a terrain classification method based on ResNet50 and transfer learning fusion is proposed to identify the terrain type in real time and improve the terrain classification accuracy. Second, a robust and precise terrain preview mapping model is developed to integrate the LiDAR with IMU data based on tightly coupled fusion and overcome the accumulation error difficulties caused by multisensor fusion. Third, introducing a decision-level fusion strategy with LiDAR and vision fusion enables the terrain classification results and three-dimensional terrain geometry information to be aligned in space and time, thereby obtaining effective terrain elevation information. As a result, the problem of inaccurate terrain elevation information can be solved. Based on the terrain detection and classification system framework, combined with the improved strategies above, the terrain classification accuracy of the training dataset and test dataset reaches 98% and 83%, respectively. The terrain mapping accuracy achieves approximately 4–5 cm in the large scene. Hence, our method effectively solves poor terrain elevation accuracy caused by the accumulated error of multisensor fusion. The experimental results highlight that the accuracy and robustness of the proposed terrain preview detection and classification algorithm in dealing with rapid vehicle rotation results are better than those of other current methods.

Building information serves as a critical foundational dataset in the fields of urban planning, smart cities and surveying and mapping, and high-resolution remote sensing (HRRS) imagery has become a vital data source for extracting building information. However, automatically extracting building information from HRRS imagery using a single feature or method remains a challenging task. On one hand, buildings exhibit significant variations in terms of their size, color, geometric structures, and other aspects. On the other hand, there are also numerous features in the environment that bear spectral and morphological resemblances to buildings. In this article, we proposed a multifeature fusion framework based on Dempster–Shafer (D-S) theory that consists of two steps for automated building extraction from HRRS imagery. The initial D-S fusion step involves two branches: object-level and pixel-level feature fusion. Then the outcomes are further combined to derive the ultimate building confidence information. In the framework, we introduced a proportional consistency and centroid consistency index to convert pixel-level features to object-level features, thereby facilitating their fusion. In addition, we proposed an initialization module for the basic probability assignment formula, enabling the elimination of the impact of nonbuilding objects and simplifying the construction process of BPAF. The experimental results based on the Nanjing, WHU, and Washington datasets demonstrate the effectiveness of our method, the accuracy outperforms the other four advanced algorithms.

Single-image rain removal is an extremely challenging task as it requires not only removing rain streaks with complex shapes, scales, and opacities but also recovering spatial details and high-level contextual structures of the underlying image. Although deep learning networks have achieved encouraging performance, current research mainly focuses on building deeper and more complex network architectures to recover reliable detailed textures or utilizing multi-scale encoder-decoder structures to learn semantic contexts in larger receptive fields, they are still not sufficient to balance rain streak removal and detail preservation. In this study, we propose a novel end-to-end network, called a dual-task complementary network (DTCN), composed of a detail recovery progressive network (DRPN) and a multifeature fusion encoder-decoder network (MEDN), to balance rain streak removal and detail preservation. Specifically, DRPN is designed to recover details in the original image, while MEDN is used to remove structural rain streaks. In addition, to reconstruct more natural and clear images, we integrate multiple network training losses, including structural similarity loss, perceptual contrast loss, perceptual image similarity loss, and edge loss. Experimental results demonstrate that our method outperforms several state-of-the-art methods on both synthetic and real rainy images. The code will be uploaded to https://github.com/zhang152267/DTCN. [ABSTRACT FROM AUTHOR]

Summary: As an important technology in computer vision, video‐based human action recognition has a great commercial value, which has attracted extensive attention in the field of computer vision and pattern recognition in both academia and industry. To date, there are a wide variety of applications of human action recognition, such as surveillance, robotics, health care, video searching, and human–computer interaction. However, there are many challenges involved in human action recognition in videos, such as cluttered backgrounds, occlusions, viewpoint variation, execution rate, and camera motion. However, data redundancy and single feature were largely limited the accuracy of human action recognition. In this article, adopting the key frame extraction and multifeature fusion techniques, a novel action recognition method was proposed, which can improve the recognition accuracy. The main works are as follows: 1) in order to solve the problem of data redundancy, a key frame extraction method based on node contribution weighting is proposed to extract video key frames; 2) different kinds of information flows are extracted from the obtained key frame sequences, and different convolutional neural networks are used to obtain corresponding classification results and merge, so as to better complement the information in different flows. Lastly, the experimental results show that our method improves the accuracy of action recognition. [ABSTRACT FROM AUTHOR]

The non-blind deblurring approach can adequately deblur single-blur images by applying a suitable mathematical model. In contrast, it cannot satisfactorily deblur images that have multiple blurs. The blind deblurring approach is able to remove various kinds of blurs from an image. However, because the causes of blur in different regions differ, it is difficult to locate and remove all the blurs accurately and also to recover the fine texture details. Considering these weaknesses and strengths of both approaches, we propose a neural network that dynamically selects suitable blur kernels for deblurring. In the proposed method, the most appropriate kernels are extracted by joint training from multiple datasets that contain specific types of blurs to tackle local and global regions in one image. In addition, to further improve the image restoration quality, we designed an edge-attention mechanism to compensate the edges and structures of specific objects. The results of experiments conducted indicate that the dynamic selection of blur kernels combined with the edge attention algorithm not only improves PSNR and SSIM, but also outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

With the advancement of Internet of Things (IoT) and artificial intelligence technologies, and the need for rapid application growth in fields, such as security entrance control and financial business trade, facial information processing has become an important means for achieving identity authentication and information security. However, in the process of acquiring facial feature information, face information is easily affected by factors, such as object occlusion, lighting changes, and similar backgrounds. In this paper, we propose a multifeature fusion algorithm based on integral histograms and a real‐time update tracking particle filtering (PF) module. First, edge features and colour features are extracted, weighting methods are used to weight the colour histogram and edge features to describe facial features, and fusion of colour features and edge features is made adaptive by using fusion coefficients to improve face tracking reliability. Then, the integral histogram is integrated into the PF algorithm to simplify the calculation steps of complex particles and improve operational efficiency. Finally, the tracking window size is adjusted in real‐time according to the change in the average distance from the particle centre to the edge of the current model and the initial model to reduce the drift problem and achieve stable tracking with significant changes in the target dimension. The results show that the algorithm improves video tracking accuracy, simplifies particle operation complexity, improves the speed, and has good anti‐interference ability and robustness compared with extracting a single feature. [ABSTRACT FROM AUTHOR]

Multispectral remote sensing images are widely used for monitoring the globe. Although thin clouds can affect all optical bands, the influences of thin clouds differ with band wavelength. When processing multispectral bands at different resolutions, many methods only remove thin clouds in visible/near-infrared bands or rescale multiresolution bands to the same resolution and then process them together. The former cannot make full use of multispectral information, and in the latter, the rescaling process will introduce noise. In this article, a deep-learning-based thin cloud removal method that fuses full spectral and spatial features in original Sentinel-2 bands is proposed, named CR4S2. A multi-input and output architecture is designed for better fusing information in all bands and reconstructing the background at original resolutions. In addition, two parallel downsampling residual blocks are designed to transfer features extracted from different depths to the bottom of the network. Experiments were conducted on a new globally distributed Sentinel-2 thin cloud removal dataset called WHUS2-CRv. The results show that the best averaged peak signal-to-noise ratio, structural similarity index measurement, normalized root-mean-square error, and spectral angle mapper of the proposed method over 12 bands in all 20 testing images were 39.55, 0.9443, 0.0245, and 2.5676°, respectively. Compared with baseline methods, the proposed CR4S2 method can better restore not only the spatial features but also spectral features. This indicates that the proposed method is very promising for removing thin clouds in multispectral remote sensing images at different resolutions.

Copyright of Journal of Zhejiang University: Science A is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Recently due to the good balance between performance and tracking speed, the discriminative correlation filter (DCF) has become a popular and excellent tracking method in short-term tracking. Computing the correlation of a response map can be efficiently performed in the Fourier domain by the discrete Fourier transform (DFT) of the input, where the DFT of an image has symmetry in the Fourier domain. However, most of the correlation filter (CF)-based trackers cannot deal with the tracking results and lack the effective mechanism to adjust the tracked errors during the tracking process, thus usually perform poorly in long-term tracking. In this paper, we propose a long-term tracking framework, which includes a tracking-by-detection part and redetection part. The tracking-by-detection part is built on a DCF framework, by integrating with a multifeature fusion model, which can effectively improve the discriminant ability of the correlation filter for some challenging situations, such as occlusion and color change. The redetection part can search the tracked object in a larger region and refine the tracking results after the tracking has failed. Benefited by the proposed redetection strategy, the tracking results are re-evaluated and refined, if it is necessary, in each frame. Moreover, the reliable estimation module in the redetection part can effectively identify whether the tracking results are correct and determine whether the redetector needs to open. The proposed redetection part utilizes a saliency detection algorithm, which is fast and valid for object detection in a limited region. These two parts can be integrated into DCF-based tracking methods to improve the long-term tracking performance and robustness. Extensive experiments on OTB2015 and VOT2016 benchmarks show that our proposed long-term tracking method has a proven effectiveness and high efficiency compared with various tracking methods. [ABSTRACT FROM AUTHOR]

In order to improve the accuracy of fire identification based on video in the Internet-of-Things environment, this article proposes a new fire identification algorithm by merging fire segmentation and multifeature fusion of fire. First, according to the relationship between R and Y channels, the improved YCbCr models are established for initial fire segmentation under reflection and nonreflection conditions, respectively. Simultaneously, the reflection and nonreflection conditions are judged by comparing the areas obtained by the two improved YCbCr models. Second, an improved region growing algorithm is proposed for fine fire segmentation by making use of the relationship between the seed point and its adjacent points. The seed points are determined using the weighted average of centroid coordinates of each segmented image. Finally, the quantitative indicators of fire identification are given according to the variation coefficient of fire area, the dispersion of centroid, and the circularity. Extensive experiments were conducted, and the experimental results demonstrate that the proposed fire detection method considerably outperforms the traditional methods on average in terms of three performance indexes: precision, recall, and $F1$ -score. Specifically, compared with the deep learning method, the precision of the proposed method is slightly higher. Although the recall of the proposed method is slightly lower than the deep learning method, its computation complexity is low. [ABSTRACT FROM AUTHOR]

This letter proposes a multifeature fusion model using deep convolution neural networks and transfer learning approach for earthquake event classification. There are several feature representations for seismic analysis, such as the time domain, the frequency domain, and the time–frequency domain. To successfully classify various earthquake events, we propose a novel model that combines these features hierarchically. In addition, we apply a transfer learning to mitigate overfitting problem of deep learning model while achieving high classification performance. To evaluate our approach, we conduct experiments with the Korean peninsula earthquake database from 2016 to 2018 and a large earthquake database on the Circum-Pacific belt in 2019. The experimental results show that the proposed method outperforms over the compared state-of-the-art methods. [ABSTRACT FROM AUTHOR]

• A multifeature fusion model was proposed for online surface roughness measurement of cold-rolled strip steel. • The statistical and depth features of speckle images were extracted to evaluate the surface roughness of cold-rolled strip steel. • Convolutional neural networks and an attention mechanism module were adopted to extract the depth features of the speckle images. • A support vector regression algorithm was used to predict the surface roughness. The online measurement of the surface roughness of cold-rolled strip steel plays a significant role in the steel manufacturing process. However, the traditional mechanism method based on laser speckle is not sufficient for image processing, and the precision of online measurement is not high. Therefore, this paper proposes a multifeature fusion model for the surface roughness of cold-rolled strip steel to improve the measurement efficiency of existing speckle methods. Cold-rolled strip steel is irradiated by a laser beam to produce speckle images. Statistical features are extracted using a gray-level cooccurrence matrix (GLCM). The deep information of an image is extracted using convolutional neural networks (CNN) and a convolutional block attention mechanism module (CBAM). The features extracted from the GLCM and CNN-CBAM methods are combined to create a multifeature fusion dataset. The fused features are predicted by a support vector regression (SVR) algorithm. Then, the SVR model is compared with three machine learning prediction models: classification and regression tree (CART), random forest (RF), and k-nearest neighbor (KNN). Experimental results confirm that the proposed multifeature fusion model predicts surface roughness with a minimum mean square error of 0.237%. [ABSTRACT FROM AUTHOR]

Light is scattered and absorbed when it is transmitted through seawater. Underwater images captured in water suffer from color attenuation, low contrast, and blurred details. To address these effects, we propose a multifeature fusion method (MFFM) to improve the color distortion and low contrast of underwater images. White balance technology is used to address the color cast of underwater images and obtain a color correction image. On this basis, the color correction image is enhanced by a guided filter to obtain a contrast-enhanced image. Four feature weights of the input images are calculated separately, and two normalized weight maps are constructed by normalization. To avoid a halo, the two normalized weight maps are multiscale fused. To improve the edge detail, the fused image is enhanced by a color adjustment technique to obtain the final enhanced image. Real underwater images are used for simulation and application testing. Compared with the current state-of-the-art methods, our method can balance the color and contrast and improve the visibility of underwater images. Extensive qualitative and quantitative experiments verify the superiority of the MFFM. [ABSTRACT FROM AUTHOR]

Recently due to the good balance between performance and tracking speed, the discriminative correlation filter (DCF) has become a popular and excellent tracking method in short-term tracking. Computing the correlation of a response map can be efficiently performed in the Fourier domain by the discrete Fourier transform (DFT) of the input, where the DFT of an image has symmetry in the Fourier domain. However, most of the correlation filter (CF)-based trackers cannot deal with the tracking results and lack the effective mechanism to adjust the tracked errors during the tracking process, thus usually perform poorly in long-term tracking. In this paper, we propose a long-term tracking framework, which includes a tracking-by-detection part and redetection part. The tracking-by-detection part is built on a DCF framework, by integrating with a multifeature fusion model, which can effectively improve the discriminant ability of the correlation filter for some challenging situations, such as occlusion and color change. The redetection part can search the tracked object in a larger region and refine the tracking results after the tracking has failed. Benefited by the proposed redetection strategy, the tracking results are re-evaluated and refined, if it is necessary, in each frame. Moreover, the reliable estimation module in the redetection part can effectively identify whether the tracking results are correct and determine whether the redetector needs to open. The proposed redetection part utilizes a saliency detection algorithm, which is fast and valid for object detection in a limited region. These two parts can be integrated into DCF-based tracking methods to improve the long-term tracking performance and robustness. Extensive experiments on OTB2015 and VOT2016 benchmarks show that our proposed long-term tracking method has a proven effectiveness and high efficiency compared with various tracking methods.

The selection and representation of remote sensing image classification features play crucial roles in image classification accuracy. To effectively improve the classification accuracy of features, an improved U-Net network framework based on multi-feature fusion perception is proposed in this paper. This framework adds the channel attention module (CAM-UNet) to the original U-Net framework and cascades the shallow features with the deep semantic features, replaces the classification layer in the original U-Net network with a support vector machine, and finally uses the majority voting game theory algorithm to fuse the multifeature classification results and obtain the final classification results. This study used the forest distribution in Xingbin District, Laibin City, Guangxi Zhuang Autonomous Region as the research object, which is based on Landsat 8 multispectral remote sensing images, and, by combining spectral features, spatial features, and advanced semantic features, overcame the influence of the reduction in spatial resolution that occurs with the deepening of the network on the classification results. The experimental results showed that the improved algorithm can improve classification accuracy. Before the improvement, the overall segmentation accuracy and segmentation accuracy of the forestland increased from 90.50% to 92.82% and from 95.66% to 97.16%, respectively. The forest cover results obtained by the algorithm proposed in this paper can be used as input data for regional ecological models, which is conducive to the development of accurate and real-time vegetation growth change models.

In this paper, a novel algorithm is presented for runway detection in synthetic aperture radar images. It involves two steps: runway assessment and confirmation. In the first step, the primary runway (PR) and the auxiliary runway (AR) of the airport are assessed by using a sparse representation fusion frame. A set of residuals, for each feature from PR or AR, is first generated by performing sparse reconstructions over two training dictionaries constructed by a set of discriminative features. Based on all residuals for all types of features, two residual sequences for PR and AR are, respectively, built. To improve the assessment performance, these two residual sequences are normalized and further linearly fused. An assessment criterion is applied to the fusion result to infer an optimal target estimate. In the second step, the histogram of oriented gradient feature descriptors of PR, AR, and the entire runway region constructed by PR, AR, and taxiways are first generated. Afterward, two semantic spatial rules are developed to verify each candidate region of interest. If a perfect match is achieved, the candidate target can be confirmed. Since the PR and AR are selected based on the residual fusion related with the concatenation of multiple features, the presented algorithm has a good representation ability to the runway. By introducing semantic spatial relationships into the confirmation scheme, this algorithm can well discriminate other runway-like targets. The test results using real scene data demonstrate that the presented method has superiority to some state-of-the-art alternatives.

In the internet age, as a conceptual model of knowledge organisation, ontology has become a research hotspot. Ontology extension achieves the purpose of expanding ontology by adding new concepts and discovering the relationships between concepts in the existing ontology. In order to improve the automation and accuracy of ontology concept extraction in the Uyghur language, here the authors propose a new method to automatically extract concepts from text collection. For the characteristics of the Uyghur domain ontology concept, the text preprocessing is performed first, then the inter‐word correlation of multi‐feature fusion is calculated, such as Mi, Cd and Ea, and finally, the domain terminology and concept are automatically extracted, according to the term frequency–inverse document frequency algorithm. Experiment results show that, in terms of precision and recall rate, the multi‐feature method proposed here represents an improvement over other methods. It also proves the feasibility and effectiveness of the authors' method to extract the domain concept. [ABSTRACT FROM AUTHOR]

In order to solve the problem that radar emitter signal modulation recognition under low signal-to-noise ratio (SNR) has difficulty in selecting the parameters of the classifier and the problem of low recognition rate caused by the improper selection of features, a multi-feature fusion modulation recognition algorithm based on improved particle swarm optimisation (PSO) algorithm is proposed. Firstly, the algorithm uses Choi–Williams distribution time–frequency analysis to transform the radar signals into time–frequency images, and combines transfer learning and principal component analysis (PCA) to extract the transfer features. Then it extracts the Renyi entropy and AR model coefficients of the signals and combines with the image features to realise multi-feature fusion. Finally, the improved PSO algorithm optimises support vector machine parameters so that population converge quickly. The simulation results show that the recognition rate of this algorithm is 95.3% when the SNR is 0 dB. The number of iterations of this algorithm is small, the signal recognition rate of the system is improved, and the effectiveness of the system is improved by applying PCA.

In order to solve the problem that radar emitter signal modulation recognition under low signal-to-noise ratio (SNR) has difficulty in selecting the parameters of the classifier and the problem of low recognition rate caused by the improper selection of features, a multi-feature fusion modulation recognition algorithm based on improved particle swarm optimisation (PSO) algorithm is proposed. Firstly, the algorithm uses Choi–Williams distribution time–frequency analysis to transform the radar signals into time–frequency images, and combines transfer learning and principal component analysis (PCA) to extract the transfer features. Then it extracts the Renyi entropy and AR model coefficients of the signals and combines with the image features to realise multi-feature fusion. Finally, the improved PSO algorithm optimises support vector machine parameters so that population converge quickly. The simulation results show that the recognition rate of this algorithm is 95.3% when the SNR is 0 dB. The number of iterations of this algorithm is small, the signal recognition rate of the system is improved, and the effectiveness of the system is improved by applying PCA. [ABSTRACT FROM AUTHOR]

A belly reconstruction and measurement scheme is conducted on a shape-flexible mannequin with digital image correlation technique. We adopted an integer subpixel image matching process. First, a compound feature including Gaussian combined moment and parameters extracted from gray-level co-occurrence matrix is proposed to track an integer pixel between reference and target deformed images. Second, a mutual learning cooperative particle swarm optimization algorithm is employed to locate the subpixel precisely. Each subpopulation does its own optimization independently. The subpopulations keep information communication and knowledge sharing for co-operative evolution to enhance global searching capability. In addition, the previous optimal information from the former interest point is adopted fully in initializing the particles' positions of the next interest point, which effectively improves the speed of the convergence. Experimental results indicate that under the high measurement accuracy without any loss, the time-consumption of this scheme is significantly superior to that of the conventional method, particularly at a large number of interest points. © 2018 SPIE and IS&T [DOI: 10.1117/1.JEI.28.2.021003] [ABSTRACT FROM AUTHOR]

Extracting and recognizing complex human movements such as sign language gestures from video sequences is a challenging task. In this paper this kind of a difficult problem is approached with Indian sign language (ISL) videos. A new segmentation algorithm is developed by fusion of features from discrete wavelet transform (DWT) and local binary pattern (LBP). A 2D point cloud is formed from fused features, which represent the local hand shapes in consecutive video frames. We validate the proposed feature extraction model with state of the art features such as HOG, SIFT and SURF for each sign video on the same ANN classifier. We found that the Haar-LBP fused features represent sign video data in better manner compared to HOG, SIFT and SURF. This is due to the combination of global and local features in our proposed feature matrix. The extracted features input the artificial neural network (ANN) classifier with labels forming the corresponding words. The proposed ANN classifier is tested against state of the art classifiers such as Adaboost, support vector machine (SVM) and other ANN methods on different features extracted from the ISL dataset. The classifiers were tested for accuracy and correctness in identifying the signs. The ANN classifier that produced a recognition rate of 92.79% was obtained with maximum training instances, which was far greater than the existing works on sign language with other features and ANN classifier on our ISL dataset. [ABSTRACT FROM AUTHOR]

Oceanic phenomena detection in synthetic aperture radar (SAR) images is important in the fields of fishery, military, and oceanography. The traditional detection methods of oceanic phenomena in SAR images are based on handcrafted features and detection thresholds, which have a problem of poor generalization ability. Methods based on deep learning have good generalization ability. However, most of the deep learning methods currently applied to oceanic phenomena detection only detect one type of phenomenon. To satisfy the requirements of efficient and accurate detection of multiple information of multiple oceanic phenomena in massive SAR images, this paper proposes an oceanic phenomena detection method in SAR images based on convolutional neural network (CNN). The method first uses ResNet-50 to extract multilevel features. Second, it uses the atrous spatial pyramid pooling (ASPP) module to extract multiscale features. Finally, it fuses multilevel features and multiscale features to detect oceanic phenomena. The SAR images acquired from the Sentinel-1 satellite are used to establish a sample dataset of oceanic phenomena. The method proposed can achieve 91% accuracy on the dataset.

Much existing work of multifeature learning relies on the agreement among different feature types to improve the clustering or classification performance. However, as different feature types could have different data characteristics, such a forced agreement among different feature types may not bring a satisfactory result. We propose a novel transductive learning approach that considers multiple feature types simultaneously to improve the classification performance. Instead of forcing different feature types to agree with each other, we perform spectral clustering in different feature types separately. Each data sample is then described by a co-occurrence of feature patterns among different feature types, and we apply these feature co-occurrence representations to perform transductive learning, such that data samples of similar feature co-occurrence pattern will share the same label. As the spectral clustering results in different feature types and the formed co-occurrence patterns influence each other under the transductive learning formulation, an iterative optimization approach is proposed to decouple these factors. Different from co-training that need to iteratively update individual feature type, our method allows all feature types to collaborate simultaneously. It can naturally handle multiple feature types together and is less sensitive to noisy feature types. The experimental results on synthetic, object, and action recognition datasets all validate the advantages of our method compared to state-of-the-art methods. [ABSTRACT FROM PUBLISHER]

In this paper, we present a novel compositional hierarchical framework for road scene understanding that allows for reliable estimation of scene topologies, such as the number, location, and width of lanes and the lane topology, i.e., parallel, splitting, or merging. In our approach, lanes and roads are represented in a hierarchical compositional model in which nodes represent parts of roads and edges represent probabilistic constraints between pairs of parts. A key benefit of our approach is the representation of lanes and roads as a set of common parts. This makes our approach applicable to scenes with rich topological diversity, while bringing along the much desired computational efficiency. To cope with the high-dimensional and continuous parameter space of our model and the non-Gaussian image evidence, we perform inference using nonparametric belief propagation. Based on this approximate inference algorithm, we introduce depth-first message passing for lane detection, which performs inference in several sweeps. Empirical results show that depth-first message passing requires significantly lower computation for performance comparable with classical belief propagation. [ABSTRACT FROM PUBLISHER]

The selection and representation of remote sensing image classification features play crucial roles in image classification accuracy. To effectively improve the classification accuracy of features, an improved U-Net network framework based on multi-feature fusion perception is proposed in this paper. This framework adds the channel attention module (CAM-UNet) to the original U-Net framework and cascades the shallow features with the deep semantic features, replaces the classification layer in the original U-Net network with a support vector machine, and finally uses the majority voting game theory algorithm to fuse the multifeature classification results and obtain the final classification results. This study used the forest distribution in Xingbin District, Laibin City, Guangxi Zhuang Autonomous Region as the research object, which is based on Landsat 8 multispectral remote sensing images, and, by combining spectral features, spatial features, and advanced semantic features, overcame the influence of the reduction in spatial resolution that occurs with the deepening of the network on the classification results. The experimental results showed that the improved algorithm can improve classification accuracy. Before the improvement, the overall segmentation accuracy and segmentation accuracy of the forestland increased from 90.50% to 92.82% and from 95.66% to 97.16%, respectively. The forest cover results obtained by the algorithm proposed in this paper can be used as input data for regional ecological models, which is conducive to the development of accurate and real-time vegetation growth change models. [ABSTRACT FROM AUTHOR]

This article introduces an image classification approach in which the semantic context of images and multiple low-level visual features are jointly exploited. The context consists of a set of semantic terms defining the classes to be associated to unclassified images. Initially, a multiobjective optimization technique is used to define a multifeature fusion model for each semantic class. Then, a Bayesian learning procedure is applied to derive a context model representing relationships among semantic classes. Finally, this context model is used to infer object classes within images. Selected results from a comprehensive experimental evaluation are reported to show the effectiveness of the proposed approaches. [ABSTRACT FROM AUTHOR]

This paper introduces an adaptive method for detecting front vehicles under complex weather conditions. In the field of vehicle detection from images extracted by cameras installed in vehicles, backgrounds with complicated weather, such as rainy and snowy days, increase the difficulty of target detection. In order to improve the accuracy and robustness of vehicle detection in front of driverless cars, a cascade vehicle detection method combining multifeature fusion and convolutional neural network (CNN) is proposed in this paper. Firstly, local binary patterns, Haar-like and orientation gradient histogram features from the front vehicle are extracted, then principal-component-analysis dimension reduction and serial-fusion processing are performed on the input image. Furthermore, a preliminary screening is conducted as the input of a support vector machine classifier based on the acquired fusion features, and the CNN model is employed to validate cascade detection of the filtered results. Finally, an integrated data set extracted from BDD, Udacity, and other data sets is utilized to test the method proposed. The recall rate is 98.69%, which is better than the traditional feature algorithm, and the recall rate of 97.32% in a complex driving environment indicates that the algorithm possesses good robustness. [ABSTRACT FROM AUTHOR]

Oceanic phenomena detection in synthetic aperture radar (SAR) images is important in the fields of fishery, military, and oceanography. The traditional detection methods of oceanic phenomena in SAR images are based on handcrafted features and detection thresholds, which have a problem of poor generalization ability. Methods based on deep learning have good generalization ability. However, most of the deep learning methods currently applied to oceanic phenomena detection only detect one type of phenomenon. To satisfy the requirements of efficient and accurate detection of multiple information of multiple oceanic phenomena in massive SAR images, this paper proposes an oceanic phenomena detection method in SAR images based on convolutional neural network (CNN). The method first uses ResNet-50 to extract multilevel features. Second, it uses the atrous spatial pyramid pooling (ASPP) module to extract multiscale features. Finally, it fuses multilevel features and multiscale features to detect oceanic phenomena. The SAR images acquired from the Sentinel-1 satellite are used to establish a sample dataset of oceanic phenomena. The method proposed can achieve 91% accuracy on the dataset. [ABSTRACT FROM AUTHOR]

Oceanic phenomena detection in synthetic aperture radar (SAR) images is important in the fields of fishery, military, and oceanography. The traditional detection methods of oceanic phenomena in SAR images are based on handcrafted features and detection thresholds, which have a problem of poor generalization ability. Methods based on deep learning have good generalization ability. However, most of the deep learning methods currently applied to oceanic phenomena detection only detect one type of phenomenon. To satisfy the requirements of efficient and accurate detection of multiple information of multiple oceanic phenomena in massive SAR images, this paper proposes an oceanic phenomena detection method in SAR images based on convolutional neural network (CNN). The method first uses ResNet-50 to extract multilevel features. Second, it uses the atrous spatial pyramid pooling (ASPP) module to extract multiscale features. Finally, it fuses multilevel features and multiscale features to detect oceanic phenomena. The SAR images acquired from the Sentinel-1 satellite are used to establish a sample dataset of oceanic phenomena. The method proposed can achieve 91% accuracy on the dataset.