Search

Under the influence of COVID-19, the economic benefits of shale gas development are greatly affected With the large-scale development and utilization of shale gas in China, it is increasingly important to assess the economic impact of shale gas development Therefore, this paper proposes a method for predicting the production of shale gas reservoirs, and uses back propagation (BP) neural network to nonlinearly fit reservoir reconstruction data to obtain shale gas well production forecasting models Experiments show that compared with the traditional BP neural network, the proposed method can effectively improve the accuracy and stability of the prediction There is a nonlinear correlation between reservoir reconstruction data and gas well production, which does not apply to traditional linear prediction methods © 2020 - IOS Press and the authors All rights reserved

Vision systems are the core information collection module in outdoor industrial systems such as factory inspection robots. However, haze greatly reduces working efficiency. Existing dehazing methods have two problems—first, they are not specifically designed for the industrial systems; second, these methods include several assumptions in their design processes and imaging models, leading to unsatisfactory results. In this article, an approach for single image dehazing is proposed to improve the efficiency of outdoor vision-based systems. First, a novel haze imaging model is proposed based on the dichromatic atmospheric scattering model. It considers the effects of multiple scattering and involves fewer assumptions. Then a data-driven technique called sparse representation is used to solve this model. Considering a haze image, a distorted and blurred version of a fine image, every patch is presented using dedicatedly prepared over-complete dictionaries and is traced back to a haze-free image. Quantitative and qualitative comparisons on a number of real-world haze images demonstrate that the proposed approach not only is more stable but also leads to better dehazing results.

Target tracking in low-altitude Unmanned Aerial Vehicle (UAV) videos faces many technical challenges due to the relatively small sizes, various orientation changes of the objects and diverse scenes. As a result, the tracking performance is still not satisfactory. In this paper, we propose a real-time single-target tracking method with multiple Region Proposal Networks (RPNs) and Distance-Intersection-over-Union (Distance-IoU) Discriminative Network (DIDNet), namely MultiRPN-DIDNet, in which ResNet50 is used as the backbone network for feature extraction. Firstly, an instance-based RPN suitable for the target tracking task is constructed under the framework of Simases Neural Network. RPN is to perform bounding box regression and classification, in which channel attention mechanism is integrated to improve the representative capability of the deep features. The RPNs built on the Block 2, Block 3 and Block 4 of ResNet50 output their own Regression (Reg) coefficients and Classification scores (Cls) respectively, which are weighted and then fused to determine the high-quality region proposals. Secondly, a DIDNet is designed to correct the candidate target’s bounding box finely through the fusion of multi-layer features, which is trained with the Distance-IoU loss. Experimental results on the public datasets of UAV20L and DTB70 show that, compared with the state-of-the-art UAV trackers, the proposed MultiRPN-DIDNet can obtain better tracking performance with fewer region proposals and correction iterations. As a result, the tracking speed has reached 33.9 frames per second (FPS), which can meet the requirements of real-time tracking tasks.

Many pieces of research have been conducted on image-restoration techniques to recover high-quality images from their low-quality versions, but they usually aim to handle a single degraded factor. However, captured images usually suffer from various degradation factors, such as low resolution and compression distortion, in the procedures of image acquisition, compression, and transmission simultaneously. Ignoring the correlation of different degraded factors may result in the limited efficiency of the existing image-restoration methods for captured images. A joint deep-network-based image-restoration algorithm is proposed to establish a restoration framework for image deblocking and super-resolution. The proposed convolutional neural network is made up of two stages. A deblocking network is constructed with two cascade deblocking subnets first, then, super-resolution is performed by a very deep network with skipping links. Cascading these two stages forms a novel deep network. An end-to-end training scheme is developed, which makes the two stages be trained jointly so as to achieve better performance. Intensive evaluations have been conducted to measure the performance of the authors’ method both in general images and face images. Experimental results on several datasets demonstrate that the proposed method outperforms other state-of-the-art methods, in terms of both subjective and objective performances.

Images/videos captured in low-light conditions often present low luminance and contrast. Although the existing low-light enhancement algorithms can improve the subjective perception, color distortion and over-enhancement are extremely obvious, which will disturb the subsequent intelligent analysis. Therefore, a naturalness-preserved low-light enhancement algorithm for intelligent analysis is proposed in this paper. An enhancement model is established in RGB color space. Images of ColorChecker color chart are captured under a series of light conditions. To preserve the naturalness, the factors of the proposed enhancement model are estimated by the images captured in practical illumination environment. Experimental results demonstrate that the proposed algorithm can produce natural enhanced results and improve the performance of vehicle license plate localization and skin color detection compared to the existing algorithms. Furthermore, the proposed algorithm can process the 720p videos at the speed of 28.3 fps on average.

Visual object tracking in unmanned aerial vehicle (UAV) videos plays an important role in a variety of fields, such as traffic data collection, traffic monitoring, as well as film and television shooting. However, it is still challenging to track the target robustly in UAV vision task due to several factors such as appearance variation, background clutter, and severe occlusion. In this paper, we propose a novel two-stage UAV tracking framework, which includes a target detection stage based on multifeature discrimination and a bounding-box estimation stage based on the instance-aware attention network. In the target detection stage, we explore a feature representation scheme for a small target that integrates handcrafted features, low-level deep features, and high-level deep features. Then, the correlation filter is used to roughly predict target location. In the bounding-box estimation stage, an instance-aware intersection over union (IoU)-Net is integrated together with an instance-aware attention network to estimate the target size based on the bounding-box proposals generated in the target detection stage. Extensive experimental results on the UAV123 and UAVDT datasets show that our tracker, running at over 25 frames per second (FPS), has superior performance as compared with state-of-the-art UAV visual tracking approaches.

DASH (Dynamic Adaptive Streaming over HTTP (HyperText Transfer Protocol)) as a universal unified multimedia streaming standard selects the appropriate video bitrate to improve the user&rsquo, s Quality of Experience (QoE) according to network conditions, client status, etc. Considering that the quantitative expression of the user&rsquo, s QoE is also a difficult point in itself, this paper researched the distortion caused due to video compression, network transmission and other aspects, and then proposes a video QoE metric for dynamic adaptive streaming services. Three-Dimensional Convolutional Neural Networks (3D CNN) and Long Short-Term Memory (LSTM) are used together to extract the deep spatial-temporal features to represent the content characteristics of the video. While accounting for the fluctuation in the quality of a video caused by bitrate switching on the QoE, other factors such as video content characteristics, video quality and video fluency, are combined to form the input feature vector. The ridge regression method is adopted to establish a QoE metric that enables to dynamically describe the relationship between the input feature vector and the value of the Mean Opinion Score (MOS). The experimental results on different datasets demonstrate that the prediction accuracy of the proposed method can achieve superior performance over the state-of-the-art methods, which proves the proposed QoE model can effectively guide the client&rsquo, s bitrate selection in dynamic adaptive streaming media services.

Estimating 3D hand pose from a single RGB image is a challenging task because of its ill-posed nature (i.e., depth ambiguity). Recently, various generative approaches have been proposed to predict the 3D joints of an RGB hand image by learning a unified latent space between two modalities (i.e., RGB image and 3D joints). However, projecting multi-modal data (i.e., RGB images and 3D joints) into a unified latent space is difficult as the modality-specific features usually interfere the learning of the optimal latent space. Hence in this paper, we propose to disentangle the latent space into two sub-latent spaces: modality- specific latent space and pose-specific latent space for 3D hand pose estimation. Our proposed method, namely Disentangled Cross-Modal Latent Space (DCMLS), consists of two variational autoencoder networks and auxiliary components which connect the two VAEs to align underlying hand poses and transfer modality-specific context from RGB to 3D. For the hand pose latent space, we align it with the two modalities by using a cross-modal discriminator with an adversarial learning strategy. For the context latent space, we learn a context translator to gain access to the cross-modal context. Experimental results on two widely used public benchmark datasets RHD and STB demonstrate that our proposed DCMLS method is able to clearly outperform the state-of-the-art ones on single image based 3D hand pose estimation.

In this paper, a vehicle color recognition method using lightweight convolutional neural network (CNN) is proposed. Firstly, a lightweight CNN network architecture is specifically designed for the recognition task, which contains five layers, i.e. three convolutional layers, a global pooling layer and a fully connected layer. Different from the existing CNN based methods that only use the features output from the final layer for recognition, in this paper, the feature maps of intermediate convolutional layers are all applied for recognition based on the fact that these convolutional features can provide hierarchical representations of the images. Spatial Pyramid Matching (SPM) strategy is adopted to divide the feature map, and each SPM sub-region is encoded to generate a feature representation vector. These feature representation vectors of convolutional layers and the output feature vector of the global pooling layer are normalized and cascaded as a whole feature vector, which is finally utilized to train Support Vector Machine classifier to obtain the recognition model. The experimental results show that, compared with the state-of-art methods, the proposed method can obtain more than 0.7% higher recognition accuracy, up to 95.41%, while the dimensionality of the feature vector is only 18% and the memory footprint is only 0.5%.

With the rapid development of Internet, the views of the online video have increased dramatically. Meanwhile, the corresponding online video advertising market showed a momentum of rapid and sustained development. In order to attract more potential purchasers and reduce the interference on the ordinary video browsers, many researchers and enterprises have conducted the research of video online advertising. At present, the insertion methods of most video advertising are always position-fixed, mandatory timing, quantitative, and the relevance of advertisement content and the video content is usually ignored. These methods will inevitably reduce the advertising effect because of browsers’ dissatisfaction and resistance. In order to overcome the shortages of the existing methods of video advertisement insertion, this paper proposed an effective content-targeted method for online video advertising. The insertion of advertisement is determined by comparing the content of the videos and the advertisements. At the same time, the characteristics of the scene switching in the video are taken into account to select the appropriate position of the advertisement insertion. Experimental results show that our method can provide a better user experience than existing methods, and its attractiveness and comfortableness is greatly improved.

Vehicle classification plays an important role in intelligent transport system. However, because the conventional vehicle classification methods are not robust to variations such as illumination, weather, noise, and the classification accuracy cannot meet the requirements of practical applications. Therefore, a new vehicle classification method using Convolutional Neural Networks is proposed in this paper, which consists of two steps: pre-training and fine-tuning. In pre-training, GoogLeNet is pre-trained on ILSVRC-2012 dataset to obtain the initial model with the corresponding connection weights. In fine-tuning, the initial model is further fine-tuned on VehicleDataset which is constructed with 13,700 images in this paper to obtain the final classification model. All images in the VehicleDataset are extracted from real highway surveillance videos, including variations of illumination, noise, resolution, angle of video cameras and weather. The vehicles are divided into six categories, i.e., bus, car, motorcycle, minibus, truck and van. The performance evaluation is carried out on the VehicleDataset. The experimental results show that the proposed method can avoid the complicated process of manually extracting features and the average classification accuracy is up to 98.26%, which is 3.42% higher than the conventional methods using “Feature + Classifier”.

In this study, a novel disturbance observer-based (DOB) [inline-formula] robust anti-disturbance tracking control problem is discussed for a kind of single-link flexible-joint manipulator with separate types of disturbances. On the one hand, a disturbance observer is imported to estimate the modelable disturbances. On the other hand, the introduction of [inline-formula] performance index optimises the impact of unmodelled disturbances on the system. A DOB proportional-integral control input is proposed such that the stability of single-link flexible-joint manipulator can be ensured. Furthermore, the satisfactory disturbance rejection and attenuation performance with [inline-formula] index can be obtained simultaneously. Finally, the simulation results verify this feasibility and effectiveness of the proposed control method.

Pedestrian attributes (such as gender, age, hairstyle, and clothing) can effectively represent the appearance of pedestrians. These are high-level semantic features that are robust to illumination, deformation, etc. Therefore, they can be widely used in person re-identification, video structuring analysis and other applications. In this paper, a pedestrian attributes recognition method for surveillance scenarios using a multi-task lightweight convolutional neural network is proposed. Firstly, the labels of the attributes for each pedestrian image are integrated into a label vector. Then, a multi-task lightweight Convolutional Neural Network (CNN) is designed, which consists of five convolutional layers, three pooling layers and two fully connected layers to extract the deep features of pedestrian images. Considering that the data distribution of the datasets is unbalanced, the loss function is improved based on the sigmoid cross-entropy, and the scale factor is added to balance the amount of various attributes data. Through training the network, the mapping relationship model between the deep features of pedestrian images and the integration label vector of their attributes is established, which can be used to predict each attribute of the pedestrian. The experiments were conducted on two public pedestrian attributes datasets in surveillance scenarios, namely PETA and RAP. The results show that, compared with the state-of-the-art pedestrian attributes recognition methods, the proposed method can achieve a superior accuracy by 91.88% on PETA and 87.44% on RAP respectively.

The limited resolution of cameras and the wide field of the video surveillance systems lead to low quality captured facial images and difficult to identify. Face super-resolution methods are proposed to enhance the resolution of facial images. However, it remains a challenging issue to restore discriminative features to identify a specific person in surveillance videos. An algorithm that helps face super-resolution and recognition with the aid of discriminative-attributes is proposed in this paper. We introduce discriminative-attributes for face recognition to recover discriminative features in the reconstructed facial images. Attributes with more discriminative power are selected to input the network together with the low-resolution face image. The experimental results of the LFW-a benchmark test show that our method achieves promising results in both subjective visual quality and face recognition accuracy.

In the preservation of old movies, there will be various degradations, such as scratches and dust. Although there have been many studies on denoising, those methods have not been effective in the restoration of old movies. In this paper, we propose a novel old film speckle noise removal method, which includes speckle noise detection and an inpainting based speckle noise reduction procedures. The spatiotemporal information is employed in speckle noise detection. Meanwhile, a deep image prior based deep auto-encoder-decoder network is designed to remove the speckle noise. Experiment results show that the proposed method can achieve better results compared with other methods.

Visual tracking is a fundamental topic in computer vision and its core issue is how to accurately and efficiently locate the object in challenging scenarios. In online tracking, improving the target positioning accuracy every frame and reducing the accumulation of tracking errors are the essential goals to obtaining robust performance. Convolutional Neural Network (CNN) based trackers have shown favorable performance gains in the problem of visual tracking in recent benchmarks. In this paper, we propose a novel bounding box adjustment module for online visual tracking. By using a lightweight Siamese neural network cascading behind a universal tracker, the confidence of the target region on two dimensions can be estimated refer to the reference image, which allow accurate adjustment of the object bounding box. Experiments are performed on OTB-2013 challenging benchmark tracking datasets and the results achieves improvement.

Fine-grained vehicle recognition plays an important part in applications, such as urban traffic management, public security, and criminal investigation. It has great chanllengs due to the subtle differences among numerous subcategories. In this paper, a fine-grained vehicle recognition method using lightweight convolutional neural network with combined learning strategy is proposed. Firstly, a lightweight Convolutional Neural Network (LWCNN) is designed specially for the fine-grained vehicle recognition task. Then, a combined training strategy, including pre-training, fine-tuning training and transfer training, is proposed to optimize the LWCNN parameters. In the pre-training phase, ILSVRC-2012 dataset is adopted to train the VGG16-Net, generating an initial model. Then, in the fine-tuning phase, the vehicle dataset is used for fine-tuning the pre-trained model to avoid learning parameters from scratch. Finally, in the transfer training phase, appropriate initialization parameters of LWCNN are obtained through the analysis of the fine-tuned network parameters. LWCNN is then trained using the vehicle dataset to obtain the highly accurate and robust classification model. Compared with the state-of-the-art methods, the proposed method can effectively decrease the computational complexity while maintaining the recognition performance.

Focusing on the overlapping areas of the logistics industry with a large number of mature blockchain applications and the public welfare industry that requires high transparency and credibility, this paper designs and implements an innovative philanthropy logistics platform based on blockchain technology through the Ethereum platform. Our platform makes use of the open, transparent, and irrevocable features of the blockchain, combined with a unique Responsibility Relay System and Evaluation and Reporting Mechanism, and can achieve the consistency of the data on the chain with real-world status, as well as the authenticity and transparency of philanthropy logistics data. This paper also establishes a model for evaluating philanthropy material donations for social welfare based on the classic network maximum flow algorithm. After four months of empirical analysis, we have concluded that the blockchain platform can greatly increase the user's trust in the project, enhance the system's cleanliness coefficient and increase the quality of philanthropically raised materials, thereby improving the public welfare of charitable donations. The paper draws the conclusion that this blockchain platform is a technical solution for maximizing social welfare.

978-1-5386-4673-1/18/$31.00 ©2018 IEEEText images in complex scenes are often suffer from various noises, which result in quality degradation and further effect the performance of automatic analysis. A novel denoise method for scene text images is proposed in this paper. With an image layered decomposition model, text images can be decomposed into a structural layer and a textual layer. Text component can be represented by the structural layer, and the texture layer of the image is closely related to the fine details and random noises. Based on this observation, a L 0 gradient minimization scheme is applied on the textual layer in our method. The L 0 gradient minimization is used to control the number of nonzero gradient values in a textured layer image, which maintains important features in the image while effectively suppress the gradients with small magnitude, which usually associate with noise. Comprehensive experimental results show that the proposed method can effectively remove the noise in the images and improve the visual quality of the text images.

Hyperspectral images are one of the most important fundamental and strategic information resources, imaging the same ground object with hundreds of spectral bands varying from the ultraviolet to the microwave. With the emergence of huge volumes of high-resolution hyperspectral images produced by all sorts of imaging sensors, processing and analysis of these images requires effective retrieval techniques. How to ensure retrieval accuracy and efficiency is a challenging task in the field of hyperspectral image retrieval. In this paper, an efficient hyperspectral image retrieval method is proposed. In principle, our method includes the following steps: (1) in order to make powerful representations for hyperspectral images, deep spectral-spatial features are extracted with the Deep Convolutional Generative Adversarial Networks (DCGAN) model; (2) considering the higher dimensionality of deep spectral-spatial features, t-Distributed Stochastic Neighbor Embedding-based Nonlinear Manifold (t-SNE-based NM) hashing is utilized to make dimensionality reduction by learning compact binary codes embedded on the intrinsic manifolds of deep spectral-spatial features for balancing between learning efficiency and retrieval accuracy; and (3) multi-index hashing in Hamming space is measured to find similar hyperspectral images. Five comparative experiments are conducted to verify the effectiveness of deep spectral-spatial features, dimensionality reduction of t-SNE-based NM hashing, and similarity measurement of multi-index hashing. The experimental results using NASA datasets show that our hyperspectral image retrieval method can achieve comparable and superior performance with less computational time.

With the rapid development of remote sensing technology, hyperspectral remote sensing image as foundation data containing abundant sensitive information has been widely applied in many fields, such as agriculture, resources, ocean, city, and environment, etc. A hybrid domain encryption method is proposed for securely transmitting and storing hyperspectral remote sensing images. Considering the spatial and spectral characteristics, the hyperspectral image is encrypted in hybrid domain (spatial and spectral). Spatial domain encryption is done by using the composite chaos sequences. Then, the spectral sequence is scrambled by the cipher sequence for protecting the spectral feature of the hyperspectral image. Finally, the spectral and spatial information is mixed by a one-to-one mapping. Experimental results on NASA datasets show that our method can effectively protect both spectral and spatial feature of hyperspectral image compared with the other methods.

Extreme weather recognition using GoogLeNet can achieve excellent performance, which is far superior to the conventional methods. However, the complexity of GoogLeNet is relatively high. Furthermore, for the small scale data, GoogLeNet usually cannot achieve the performance as the large scale data does. In this paper, a novel dual fine-tuning strategy is proposed to train the GoogLeNet model. Firstly, ILSVRC-2012 Dataset is applied to pre- train GoogLeNet and the initial model can be generated. Then, the initial model is fine-tuned on WeatherDataset, which is constructed in this paper, to obtain the model-1 for extreme weather recognition. Next, the structure of GoogLeNet is optimized by truncating operation to reduce the model size of GoogLeNet. And the truncated GoogLeNet is further fine-tuned on WeatherDataset to obtain the final recognition model. The experimental results have demonstrated that, compared with the original GoogLeNet, the recognition accuracy of the proposed method increases from 94.74% to 95.46%, which is improved by 0.72%. The model size of the optimized GoogLeNet is only 31.23% of original GoogLeNet. On CPU and GPU implementation, the proposed method processes the images 1.39 times faster and 2.44 faster than the original GoogLeNet respectively. In summary, the proposed method is better than the original GoogLeNet in the three aspects of recognition accuracy, recognition speed and model size.

The hyperspectral images are represented as an image cube, which image the same ground object with several tens to hundreds of spectral bands from ultraviolet to the microwave range. Content-Based Image Retrieval (CBIR) for hyperspectral images has been explored in recent years. However, it is still a concern and challenging task to extract a highly descriptive feature to contribute to improve the performance of hyperspectral image retrieval. At present, deep learning is a new research area of machine learning, which can extract more effective deep features by using a cascade of many layers of nonlinear processing units. In this paper, a deep spectral-spatial feature extraction method is proposed based on Deep Convolutional Generative Adversarial Networks (DCGAN) for hyperspectral image retrieval. Firstly, spectral vector is extracted by selecting manually pure pixels from hyperspectral image. Then spatial vector is extracted by selecting manually the neighbor pixels of pure pixels from principal bands after reducing the dimensionality of hyperspectral images with one-bit (1BT) transform. Spectral-spatial vector as training samples is obtained by combining spectral vector with spatial vector by using Vector Stacking (VS) approach. After training DCGAN model, deep spectral-spatial feature is extracted to further apply into hyperspectral retrieval. Experiments are conducted among our method and the three other state-of-the-art methods including endmember extraction using improved Automatic Pixel Purity Index (APPI), spectral and spatial features extraction, and endmember signatures extraction using Endmember Induction Algorithm (EIA). Experimental results on AVIRIS data show that our method can achieve a higher accuracy for hyperspectral image retrieval and further prove our extracted deep spectral-spatial feature has stronger descriptive ability.

The tags are usually tagged by different users in social image sharing websites, which can indicate image semantic information and imply user's preference. Therefore, the tags can contribute to personalized recommendation of social image. However, the present social image tags models only consider single tag, resulting in the relationships among tags are ignored. In this paper, we propose a novel method to create tag tree of social image for personalized recommendation. Firstly, the tag ranking is realized to remove noisy tags. Then, the first layer tags are selected from re-ranked tags lists. To sufficiently express tag's significances, the tag subtrees can be created based on different image categories and combined with first layer tags to create tag tree. Finally, the personalized recommendation of social image is achieved by using tag tree. Experimental results show that our tag tree can effectively express the relationships among tags as well as obtain satisfactory results in personalized recommendation of social image.

In the procedures of image acquisition, compression, and transmission, captured images usually suffer from various degradations, such as low-resolution and compression distortion. Although there have been a lot of research done on image restoration, they usually aim to deal with a single degraded factor, ignoring the correlation of different degradations. To establish a restoration framework for multiple degradations, a joint deep-network-based image restoration algorithm is proposed in this paper. The proposed convolutional neural network is composed of two stages. Firstly, a de-blocking subnet is constructed, using two cascaded neural network. Then, super-resolution is carried out by a 20-layer very deep network with skipping links. Cascading these two stages forms a novel deep network. Experimental results on the Set5, Setl4 and BSD100 benchmarks demonstrate that the proposed method can achieve better results, in terms of both the subjective and objective performances.

Person re-identification (re-id) aims to match a specific person across non-overlapping views of different cameras, which is currently one of the hot topics in computer vision. Compared with image-based person re-id, video-based techniques could achieve better performance by fully utilizing the space-time information. This paper presents a novel video-based person re-id method named Deep Feature Guided Pooling (DFGP), which can take full advantage of the space-time information. The contributions of the method are in the following aspects: (1) PCA-based convolutional network (PCN), a lightweight deep learning network, is trained to generate deep features of video frames. Deep features are aggregated by average pooling to obtain person deep feature vectors. The vectors are utilized to guide the generation of human appearance features, which makes the appearance features robust to the severe noise in videos. (2) Hand-crafted local features of videos are aggregated by max pooling to reinforce the motion variations of different persons. In this way, the human descriptors are more discriminative. (3) The final human descriptors are composed of deep features and hand-crafted local features to take their own advantages and the performance of identification is promoted. Experimental results show that our approach outperforms six other state-of-the-art video-based methods on the challenging PRID 2011 and iLIDS-VID video-based person re-id datasets.

Color is one of the most stable attributes of vehicles and often used as a valuable cue in some important applications. Various complex environmental factors, such as illumination, weather, noise and etc., result in the visual characteristics of the vehicle color being obvious diversity. Vehicle color recognition in complex environments has been a challenging task. The state-of-the-arts methods roughly take the whole image for color recognition, but many parts of the images such as car windows; wheels and background contain no color information, which will have negative impact on the recognition accuracy. In this paper, a novel vehicle color recognition method using local vehicle-color saliency detection and dual-orientational dimensionality reduction of convolutional neural network (CNN) deep features has been proposed. The novelty of the proposed method includes two parts: (1) a local vehicle-color saliency detection method has been proposed to determine the vehicle color region of the vehicle image and exclude the influence of non-color regions on the recognition accuracy; (2) dual-orientational dimensionality reduction strategy has been designed to greatly reduce the dimensionality of deep features that are learnt from CNN, which will greatly mitigate the storage and computational burden of the subsequent processing, while improving the recognition accuracy. Furthermore, linear support vector machine is adopted as the classifier to train the dimensionality reduced features to obtain the recognition model. The experimental results on public dataset demonstrate that the proposed method can achieve superior recognition performance over the state-of-the-arts methods.

With the explosive growth of mobile videos, helping users quickly and effectively find mobile videos of interest and further provide personalized recommendation services are the developing trends of mobile video applications. Mobile videos are characterized by their wide variety, single content, and short duration, and thus traditional personalized video recommendation methods cannot produce effective recommendation performance. Therefore, a personalized mobile video recommendation method is proposed based on user preference modeling by deep features and social tags. The main contribution of our work is three-fold: (1) deep features of mobile videos are extracted by an improved exponential linear units-3D convolutional neural network (ELU-3DCNN) for representing video content, (2) user preference is modeled by combining user preference for deep features with user preference for social tags that are respectively modeled by maximum likelihood estimation and exponential moving average method, (3) a personalized mobile video recommendation system based on user preference modeling is built after detecting key frames with a differential evolution optimization algorithm. Experiments on YouTube-8M dataset have shown that our method outperforms state-of-the-art methods in terms of both precision and recall of personalized mobile video recommendation.

Vehicle classification is vital to an intelligent transport system. To obtain a high accuracy, it is the most crucial process to extract reliable and distinguishable features of vehicles. A feature extraction method using a lightweight convolutional network for vehicle classification is proposed. The main contributions are threefold: (1) a lightweight network named LWNet with two convolution layers is proposed to extract the features of the vehicles; (2) Hu moment is integrated with spatial location information to improve its own describing and distinguishing ability; and (3) histogram of oriented gradient (HOG) feature is extracted from the complete image, and then the above two kinds of features are combined with HOG to form the vector. And then, a support vector machine is trained to obtain the classification model. Vehicles are classified into six categories, i.e., large bus, car, motorcycle, minibus, truck, and van. The experimental results have demonstrated that the classification accuracy can achieve 97.39%, which is 3.81% higher than that obtained from the conventional methods. In addition, for this vehicle classification task, the proposed lightweight convolutional network can achieve comparable or even higher performance compared to the deep convolutional neural networks, while the proposed method does not need the support of a graphics processing unit and has much lower complexity without the training process.