As one of the important fields of computer vision research, pedestrian attribute recognition has gained increasing attention from domestic and foreign researchers due to its huge potential applications. However, obtaining long-distance pedestrian information in actual scenes poses challenges such as lack of information, incomplete feature extraction, and low attribute recognition accuracy. To address these issues, we propose a multi-scale feature fusion network based on a dual self-attention mechanism. The fusion module merges multi-scale features to enable more complete attribute extraction, while the dual self-attention module focuses the network on important regions. Experimental results on PA-100K, RAP, and PETA datasets achieved mean accuracies of 81.97
Aiming at the problems of complex pre-processing and expensive equipment in chemical detection of graphite ore grade, a graphite ore identification and classification method based on fusing multi-scale image features is proposed. In the feature extraction stage, a deep convolutional neural network and a residual network model based on spatial attention mechanism are constructed to improve the learning ability of local and global features of graphite ore images; in the feature aggregation stage, a global response normalization technique is introduced to achieve more accurate graphite ore grade recognition, and the accuracy of the model reaches 93.401
Human skeleton point data has better environmental adaptability and motion expression ability than RGB video data. Therefore, the action recognition algorithm based on skeletal point data has received more and more attention and research. In recent years, skeletal point action recognition models based on graph convolutional networks (GCN) have demonstrated outstanding performance. However, most GCN-based skeletal action recognition models use three stable spatial configuration partitions, and manually set the connection relationship between each skeletal joint point. Resulting in an inability to better adapt to varying characteristics of different actions. And all channels of the input X features use the same graph convolution kernel, resulting in coupling aggregation. Contrary to the above problems, this paper proposes a new division strategy, which can better extract the feature information of neighbor nodes of nodes in the skeleton graph and adaptively obtain the connection relationship of joint nodes. And introduce Decoupled Graph Convolution (DC-GCN) to each partition to solve the coupled aggregation problem. Experiments on the NTU-RGB+D dataset show that the proposed method can achieve higher action recognition accuracy than most current methods.
For the problem of generating high-quality and diverse images, an image generation method combining residual module, spectral parametric normalization, and self-attention mechanism is proposed to be applied in WGAN networks. The specific improvement of the method is to introduce the residual module into the generator and discriminator networks to better capture the deep image information. The spectral parametric normalization technique is also applied to each convolutional layer of the residual block to improve the stability of the image generation process. The self-attention mechanism is introduced into the generator to enable the network to learn in a targeted manner and generate higher-quality images. The experimental results demonstrate that the combined application of these techniques can effectively solve the challenge of generating image samples, obtain stable and diverse data samples, generate better results than the original WGAN method and DCGAN method, and use the generated data samples as the dataset for expanding the classification experiments, which improves the recognition accuracy of the image classification network to a certain extent.
Light field (LF) camera sensors often face a trade-off between angular resolution and spatial resolution when shooting. High spatial resolution image arrays often result in lower angular resolution, and vice versa. In order to obtain high spatial resolution and at the same time have high angular resolution. In this paper, we propose an improved 4D convolutional neural network (CNN) algorithm for angular super-resolution (SR) to improve the quality of angular SR images. Firstly, to address the problem of low luminance of images captured by LF cameras, this paper uses block threshold square reinforcement (BTSR) for image luminance enhancement. Secondly, to make the reconstructed new viewpoints of higher quality, this paper improves the attention mechanism convolutional block attention module (CBAM). This paper incorporates it into a 4D dense residual network as high dimensional attention module (HDAM). HDAM generates images along two independent dimensions, spatial and channel. The HDAM generates attention maps along two independent dimensions, space and channel, which guide the network to focus on more important features for adaptive feature modification. Finally, this paper modifies the activation function to make the network perform better in the later stages of training and more suitable for LF reconstruction tasks. This paper evaluates the network on many LF data, including real-world scenes and synthetic data. The experimental results show that the improved network algorithm can achieve higher quality LF reconstruction.
In recent years, numerous Deep Reinforcement Learning (DRL) neural network models have been proposed to optimize computational offloading and resource allocation in Mobile Edge Computing (MEC). However, the diversity of computational tasks and the complexity of 5G networks pose significant challenges for current DRL algorithms apply to MEC scenarios. This research focuses on a single MEC server-multi-user scenario and develops a realistic small-scale MEC offloading system. In order to alleviate the problem of overestimation of action value in current Deep Q-learning Network (DQN), we propose a normalized model of Complex network based on Double DQN (DDQN) algorithm to determine the optimal computational offloading and resource allocation strategy. Simulation results demonstrate that DDQN outperforms conventional approaches such as fixed parameter policies and DQN regarding convergence speed, energy consumption and latency. This research showcases the potential of DDQN for achieving efficient optimization in MEC environments.
Using the Transformer for motion deblurring enables a broader receptive field, and by stacking multiple Transformer modules, it captures global correlations in features. However, this increases network complexity and poses convergence challenges. To address this, a Generative Adversarial Network called XT-GAN, which combines multiple-scale Transformers, has been proposed.XT-GAN leverages pyramid features from a convolutional network as a lightweight substitute for multi-scale inputs. Within the output pyramid convolutional features, different-scale features are computed in parallel using multi-head self-attention. These features are combined with a proposed feature enhancement module to represent information at different scales. Finally, the network outputs from various modules are concatenated and restored to the original image size.In experiments conducted on the synthetic dataset GoPro, XT-GAN outperformed ordinary networks such as DeblurGAN, DeepDeblur, and SRN. It achieved a reduction in computational complexity of at least 70
Using unmanned aerial vehicles (UAV) for large-scale scene sampling is a prevalent application in UAV vision. However, there are certain factors that can influence the quality of UAV sampling, such as the lack of texture details and drastic changes in scene geometry. One common factor is occlusion, which is a surface feature in 3D scenes that results in significant discontinuity on the scene surface, leading to transient noise and loss of local information. This can cause degradation in the performance of computer vision algorithms. To address these challenges, this paper proposes a UAV sampling method that takes into account occlusion. The method is based on the principle of quantizing occlusion information and improves the aerial light field (ALF) technology. It establishes a UAV ALF sampling model that considers scene occlusion information and calculates the minimum sampling rate of UAV sampling by deriving the exact expression of the spectrum. The proposed model is used to sample and reconstruct large-scale scenes in different occlusion environments. Experimental results demonstrate that the model effectively improves the reconstruction quality of large-scale scenes in occluded environments.
Ultrasound is commonly used for diagnosis and detection in a variety of fields, and the analysis of ultrasound echo signals presents a significant challenge in terms of the amount of time required by professionals to make subjective judgements. With the advances made in artificial intelligence technology on computers, more and more fields are being aided by it, not only increasing efficiency but also improving overall accuracy. In this paper, an inverse pyramid pooling of attention (IPPA) mechanism is proposed for images transformed from ultrasound echo signals. IPPA performs different pooling operations at multiple scale levels for each channel of the feature matrix, obtaining rich regional feature associations and thus improving the representation of the channels. In addition, different probability factors were assigned for the different pooling, and domain channel information was extracted by adaptive 1D convolution to enhance the adaptation range of the network model. Experimental results on a 10-class ultrasound hyperdata set (consisting of three sub-datasets) show that the sensitivity and robustness of the ResNet integrated with IPPA are improved over the original ResNet, with an accuracy of up to 99.68%.
In the field of computer vision, pedestrian attribute recognition plays a crucial role in pedestrian detection and pedestrian re-identification. However, this task faces challenges such as blurry images, difficulty in recognizing fine-grained features, and overlooking relationships between pedestrian attributes. To address these challenges, we propose a novel method for pedestrian attribute recognition. Our method is based on convolutional neural networks and incorporates a feature pyramid structure that is specifically designed for the task of pedestrian attribute recognition (PAR). Additionally, we enhance feature information by employing multi-scale feature fusion. Furthermore, our proposed AIIM module facilitates interactions between different attributes by establishing both remote dependencies and short-range dependencies. Through comprehensive experimentation, we have validated the effectiveness of our method and achieved state-of-the-art results. Specifically, our method has achieved impressive average accuracies (mA) of 86.27
Although animation synthesis technology is widely applied, it also imposes higher demands on the precision of the synthesized animation. This paper employs a more lightweight channel attention module for image feature extraction. Compared to previous channel attention module, this approach utilizes fewer parameters, thereby assisting the network in achieving improved precision. Additionally, it replaces the sigmoid function with the more suitable output function tanh for image generation. Three evaluation metrics show improvements: a 1.3
In response to the issues of high centralization, slow transaction rates, and high security risks in the cross-chain mechanism of notary groups, this paper proposes an identity-based, non-interactive cross-chain model for notary groups. The model introduces notary groups to reduce centralization and divides the nodes within the notary group into transaction nodes, validation nodes, and supervisory nodes using verifiable random functions, significantly improving the fault tolerance of the optimized model. Additionally, the model introduces Merkle tree structures to locally store transaction information, enabling the processing of multiple cross-chain transactions at once and reducing transaction latency caused by multiple verifications, thereby improving transaction rates. Experimental results demonstrate that compared to traditional models, the optimized model significantly reduces transaction security risks and increases transaction rates by 55.8%.
Federated Learning (FL) has recently attracted considerable attention in multi-robot collaborative systems, owning to its capability of enabling mobile clients to collaboratively learn a global prediction model without sharing their privacy-sensitive data to the server. In a multi-robot collaboration system, an approach that ensures privacy-preserving knowledge sharing among multiple robots becomes imperative. However, the application of FL in such systems encounters two major challenges. Firstly, it is inefficient to use all the network nodes as federated learning clients (which conduct training of machine learning model based on own data) due to the limited wireless bandwidth and energy of robots. Secondly, the selection of an appropriate number of clients must be carefully considered, considering the constraints imposed by limited communication resources. Selecting an excessive number of clients may result in a failure in uploading important models. To overcome these challenges, this paper proposes a client selection approach that considers multiple metrics including the data volume, computational capability, and network environment by integrating fuzzy logic and Q-learning. The experimental results validate the theoretical feasibility of the proposed approach. Further empirical data can be derived from training experiments on public datasets, enhancing the practical applicability of the proposed method.
Content prediction can avoid VR video streaming delay in mobile edge caching system. To reduce request delay, popular content should be cached on edge server. Existing work either focuses on content prediction or on caching algorithms. However, in the end-edge-cloud system, prediction and caching should be considered together. In this paper, we jointly optimize the four stages of prediction, caching, computing and transmission in mobile edge caching system, aimed to maximize the user's quality of experience. We propose a progressive policy to optimize the four steps of VR video streaming. Since the user's QoE is determined by the performance of the resource allocation and caching algorithm, we design a caching algorithm with unknown future request content, which can efficiently improve the content hit rate, as well as the durations for prediction, computing and transmission. We optimize the four stages under arbitrary resource allocation and simulate the proposed algorithm according to the degree of overlap, as well as completion rate. Finally, under the real scenario, the proposed algorithm is verified by comparing with several other caching algorithms, simulation results show that the user's QoE is improved under the progressive policy and the proposed algorithm.
Knowledge graph is a large database composed of entities, relationships and attributes, which can provide rich semantic information for search engines. The potential semantic relation search engine based on Knowledge graph is a novel search engine. It obtains potential semantic relationships from the Knowledge graph, and then uses these potential semantic relationships to search for data sources such as web pages and documents. This paper first analyzes the characteristics of the Knowledge graph, then lists the construction process of the Knowledge graph based on WordNet, and finally proposes the potential semantic relationship search engine architecture based on the Knowledge graph.
To address the problems of low recognition accuracy and long training time of the original C3D (Convolutional 3D) model, this paper proposes a modified method to improve its framework. Firstly, the Relu activation function in the hidden layer is replaced by the Hardswish function to allow more neurons to participate in parameter updating and to alleviate the problem of slow gradient convergence. Secondly, the dataset was optimised using the background difference method and the image scaling improvement respectively, and the optimised dataset was used for model training. The image scaling improvement combined with the activation function improvement results in a better HDs-C3D (Hardswish Data scaling - Convolutional 3D) model. Its accuracy on the training dataset reached 89.1
Users employ cloud servers to store data and depend on third-party audits to guarantee data integrity. However, this auditing system poses certain risks, as it may have vulnerabilities that attackers can exploit for intrusion. To address these concerns and achieve decentralization, a cloud data integrity verification scheme is proposed. This scheme is based on a verifiable random function and aims to eliminate the need for third-party auditing. Before performing data integrity verification, a blockchain smart contract is employed to calculate bilinear pairs, serving the purpose of verifying the user's legitimacy. If the user successfully completes this verification, the integrity of the cloud data is then verified using the verifiable random function. The simulation results demonstrate that this scheme is effective in detecting the legitimacy of users and significantly reduces the computational and communication overhead associated with verifying data integrity.
In order to determine whether the electric power workers wear safety equipment such as safety helmet, insulation boots, insulation gloves, insulation clothes, etc., to ensure the safety of the electric power construction site. We propose a electric power operation safety equipment detection algorithm incorporating PSA to improve YOLOv5s algorithm, using polarized self-attention mechanism to improve the feature extraction end of YOLOv5s algorithm, improving the channel resolution and spatial resolution of safety equipment images of electric power operation scenes, and preserving the information of key nodes of small targets that are obscured; GSConv is used to replace the ordinary convolution to reduce the complexity of the model, improve the calculation speed of the algorithm and improve the detection accuracy. The experimental results show that the average accuracy mean (IoU = 0.5) of the proposed algorithm reaches 0.961, which is 1.58
Based on the tedious process of using a carbon-sulfur analyzer to detect the carbon grade of graphite in graphite mining production, this paper proposes a graphite carbon grade image recognition and classification method based on multi-scale feature fusion. The experiment preprocesses the images and constructs a residual network model that combines pyramid convolution (PyConv) and spatial attention mechanism (SAM). This model enhances the extraction of both global and local feature information from graphite images. Transfer learning is introduced by using pre-trained weights to accelerate the convergence of the model, achieving efficient and accurate recognition and classification of graphite carbon grade with an accuracy of 92.5
Application scenarios such as unmanned driving and UAV reconnaissance have the requirements of high performance, low delay and small space occupation. Images taken in foggy days are easy to be affected by fog or haze, thus losing some important information. The purpose of image dehazing is to remove the influence of fog on image quality, which is of great significance to assist in solving high-level vision tasks. Aiming at the shortcomings of the current defogging method, such as slow defogging speed and poor defogging effect, this paper introduces the idea of FPCNet and the attention mechanism module, and proposes an improved AODNet fast defogging algorithm to ensure the defogging speed and defogging performance. The public dataset RESIDE was used for training and testing. Experimental results show that in terms of dehazing performance, the proposed algorithm achieves 25.78 and 0.992 in PSNR and SSIM respectively. In terms of dehazing speed, the proposed method is close to AODNet, with only 5 times more parameters than AODNet, but more than 100 times smaller than other methods.
