Search

Abstract This work addresses a critical issue: the deterioration of concrete structures due to fine-grained cracks, which compromises their strength and longevity. To tackle this problem, experts have turned to computer vision (CV) based automated strategies, incorporating object detection and image segmentation techniques. Recent efforts have integrated complex techniques such as deep convolutional neural networks (DCNNs) and transformers for this task. However, these techniques encounter challenges in localizing fine-grained cracks. This paper presents a self-supervised 'you only look once' (SS-YOLO) approach that utilizes a YOLOv8 model. The novel methodology amalgamates different attention approaches and pseudo-labeling techniques, effectively addressing challenges in fine-grained crack detection and segmentation in concrete structures. It utilizes convolution block attention (CBAM) and Gaussian adaptive weight distribution multi-head self-attention (GAWD-MHSA) modules to accurately identify and segment fine-grained cracks in concrete buildings. Additionally, the assimilation of curriculum learning-based self-supervised pseudo-labeling (CL-SSPL) enhances the model's ability when applied to limited-size data. The efficacy and viability of the proposed approach are demonstrated through experimentation, results, and ablation analysis. Experimental results indicate a mean average precision (mAP) of at least 90.01%, an F1 score of 87%, and an intersection over union threshold greater than 85%. It is evident from the results that the proposed method yielded at least 2.62% and 4.40% improvement in mAP and F1 values, respectively, when tested on three diverse datasets. Moreover, the inference time taken per image is 2 ms less than that of the compared methods.

This study addresses the Weighted Multi-Skill Resource Constrained Project Scheduling Problem (W-MSRCSPSP) with the aim of minimizing software project makespan. Unlike previous works, our investigation regards heterogeneous resources characterized by varying skill proficiency levels. Another major problem with existing methodologies is the potential underutilization of human resources due to varying task durations. This work introduces an innovative scheduling approach known as the Greedy and Parallel Scheduling (GPS) algorithm to handle the said issues. GPS focuses on assigning the most suitable resources available to project activities at each scheduling point. The fundamental goal of our proposed approach is to reduce resource wastage while efficiently allocating surplus resources, if any, to project tasks, ultimately leading to a decrease in the makespan. To empirically evaluate the efficacy of the GPS algorithm, we conduct a comparative analysis against the Parallel Scheduling Scheme (PSS). The advantage of our proposed approach lies in its ability to optimize the utilization of available resources, resulting in accelerated project completion. Results from extensive simulations substantiate this claim, demonstrating that the GPS scheme outperforms the PSS approach in minimizing project duration.

Various supervised machine-learning algorithms for wind power forecasting have been developed in recent years to manage wind power fluctuations and effectively correlate to energy consumption; Meanwhile, the performance of the model does suffer from missing values. To address the issue of missing values in wind power forecast, this paper proposes two methods: Clue-based missing at random (CMAR) and patterned k-nearest Neighbor (PkNN). In addition, a hybrid wind energy forecasting system has been created that is built on 1D-Convolutional neural networks, which are used to extract features from raw input, and Long Short-Term Memory, which employs time series data internal representation learning to improve the accuracy of month-wise wind power forecasting. The efficacy of the proposed model on generated datasets is also compared to the classic machine learning model to check the generalization ability. The strength of the Convolutional LSTM has been estimated in terms of different performance metrics such as mean absolute error (MAE), root mean squared error (RMSE), R2, and explained variance score (EVS). The experimental results show that the use of the PkNN algorithm for data imputation; integrated with regression-based Convolutional LSTM is much more efficient in prediction over other deep neural network models.

Abstract Intelligent robotics and expert system applications in dentistry suffer from identification and detection problems due to the non-uniform brightness and low contrast in the captured images. Moreover, during the diagnostic process, exposure of sensitive facial parts to ionizing radiations (e.g., X-Rays) has several disadvantages and provides a limited angle for the view of vision. Capturing high-quality medical images with advanced digital devices is challenging, and processing these images distorts the contrast and visual quality. It curtails the performance of potential intelligent and expert systems and disincentives the early diagnosis of oral and dental diseases. The traditional enhancement methods are designed for specific conditions, and network-based methods rely on large-scale datasets with limited adaptability towards varying conditions. This paper proposed a novel and adaptive dental image enhancement strategy based on a small dataset and proposed a paired branch Denticle-Edification network (Ded-Net). The input dental images are decomposed into reflection and illumination in a multilayer Denticle network (De-Net). The subsequent enhancement operations are performed to remove the hidden degradation of reflection and illumination. The adaptive illumination consistency is maintained through the Edification network (Ed-Net). The network is regularized following the decomposition congruity of the input data and provides user-specific freedom of adaptability towards desired contrast levels. The experimental results demonstrate that the proposed method improves visibility and contrast and preserves the edges and boundaries of the low-contrast input images. It proves that the proposed method is suitable for intelligent and expert system applications for future dental imaging.

Target tracking is one of the challenging tasks in computer vision. Usually, the center of target origins from the position with the largest response value, and the key to improving tracking performance is to learn reliable feature maps. This paper analyzes the characteristics of the tracking task, designs a global feature comparison function to extract the context, and proposes a feature supplement module based on the global comparison information for further performance improvement. In addition, we also design a template feature update module to supplement template features based on the search area features of the current frame to dynamically adjust model features, improve model generalization capabilities, and avoid model feature fixation. The proposed feature supplement model based on global feature comparison (FSGFC) is evaluated on five visual tracking benchmarks including OTB100, VOT2016, VOT2018, VOT2019 and UAV123. The experimental results show that the model obtains the state-of-the-art performance with a real-time speed.

Alzheimer's is an acute degenerative disease affecting the elderly population all over the world. The detection of disease at an early stage in the absence of a large-scale annotated dataset is crucial to the clinical treatment for the prevention and early detection of Alzheimer's disease (AD). In this study, we propose a transfer learning base approach to classify various stages of AD. The proposed model can distinguish between normal control (NC), early mild cognitive impairment (EMCI), late mild cognitive impairment (LMCI), and AD. In this regard, we apply tissue segmentation to extract the gray matter from the MRI scans obtained from the Alzheimer's Disease National Initiative (ADNI) database. We utilize this gray matter to tune the pre-trained VGG architecture while freezing the features of the ImageNet database. It is achieved through the addition of a layer with step-wise freezing of the existing blocks in the network. It not only assists transfer learning but also contributes to learning new features efficiently. Extensive experiments are conducted and results demonstrate the superiority of the proposed approach.

Unmanned aerial vehicles (UAVs) are promising in large-area data collection due to their flexibility and easy maintenance. In this work, we study a UAV-enabled wireless sensor network (WSN), where K UAVs are dispatched to collect a certain amount of data from each node on the ground. Most existing works assume that the flight energy is either distance-related or duration-related, which may not suit the practical scenario. Given the practical speed-related flight energy model, we focus on deriving the optimal energy and delay tradeoff for the K UAVs such that each node can successfully upload a certain amount of data to one of the K UAVs. Intuitively, the higher flight speed of the UAV results in the shorter completion time of the data collection task, which may however cause the higher flight energy consumption of UAVs during the task. Specifically, we first model the total energy consumption of the UAV during the flight for collecting data within the WSN and then design the flight speed as well as the flight trajectory of each UAV for achieving different Pareto-optimal tradeoffs between the maximum single-UAV energy consumption among all UAVs and the task completion time. To achieve this goal, we propose a novel multi-objective ant colony optimization framework based on the adaptive coordinate method (MOACO-ACM). Firstly, the adaptive coordinate method is developed to decide the nodes visited by each of the K UAVs, respectively. Secondly, the ant colony algorithm is incorporated to optimize the visiting order of nodes for each UAV. Finally, we discuss the impact of UAVs’ speeds scheduling on the tradeoff between the task completion time and the maximum single-UAV energy consumption among all UAVs. Extensive simulations validate the effectiveness of our designed algorithm and further highlight the importance of UAVs’ flight speeds in achieving both energy-efficient and time-efficient data collection.

Deep hashing has received a great deal of attraction in large-scale data analysis, due to its high efficiency and effectiveness. The performance of deep hashing models heavily relies on label information, which is very expensive to obtain. In this work, a novel end-to-end deep hashing model based on pseudo-labels for large-scale data without labels is proposed. The proposed hashing model consists of two major stages, where the first stage aims to obtain pseudo-labels based on deep features extracted by a pre-training deep convolution neural network. The second stage generates hash codes with high quality by the same neural network in the previous stage, coupled with an end-to-end hash layer, whose purpose is to encode data into a binary representation. Additionally, a quantization loss is introduced and interwound within these two stages. Evaluation experiments were conducted on two frequently-used image collections, CIFAR-10 and NUS-WIDE, with eight popular shallow and deep hashing models. The experimental results show the superiority of the proposed method in image retrieval.

Aiming at alleviate the detrimental effect of similar object interferences and target state changes in SiamRPN tracker, a Channel Positive and Negative Feedback Network (CPFN) is proposed, in which the Gaussian score map is generated by the feature channels selected by a Gaussian kernel, and the map is combined with the classification branches of SiamRPN. In this way, the feature channels are divided into positive feedback channels and interference channels, and these feature channels are effectively utilized. In addition, a channel weight update strategy is proposed to enhance the robustness of the tracker and avoid template pollution caused by inadequate template update. Extensive experiments on tracking benchmarks including VOT2016, VOT2018, VOT2019, OTB100, UAV123, LaSOT and GOT-10k show that the proposed CPFN outperforms the state-of-the-art methods based on small backbone network in terms of accuracy and achieves high-speed tracking.

Image deraining is a low-level restoration task that has become quite popular during the past decades. Although recent data-driven deraining models exhibit promising results, most of these models are trained on synthetic rain data sets which do not generalize well when applied to real rain images. While recent real-rain data sets have achieved favorable generalization performance, generating rain-free ground-truths can be tedious and time-consuming. To address this problem, in this work, we present rain to rain training, an unsupervised training method for single image deraining. Our experiments show that it is possible to train single image deraining models by using only rain images. This can be achieved by simply training models to map pairs of rain images. We also introduce the idea of using the least overlapping training pairs, a method of selecting adequate training pairs that enables rain to rain training to achieve equivalent deraining performance compared to supervised training.

The main advantage of multicast over multi-unicast is the existence of transmission reuse, i.e., the cooperation among destinations. This intrinsic characteristic benefits the multicast regarding the enhancement of transmission efficiency. Consequently, multicast is extensively studied under various wireless environments. Yet people still have limited understanding on the impact of node mobility on the transmission reuse when multicast is applied instead of multiple unicast. In this paper, we focus on the correlated mobility which captures the feature of real mobility processes, to study its influence on transmission reuse in mobile ad hoc networks (MANETs). Specifically, we quantify the transmission reuse as multicast gain, i.e., the capacity ratio of multicast and multi-unicast under certain delay constraint. We design a multi-layer routing protocol and propose different kinds of causal scheduling policies, under which the overall multicast capacity-delay tradeoff is derived by exploring various correlation degrees of node mobility. Compared with the capacity-delay tradeoff in unicast case, we calculate the multicast gain. Results show that the correlation of node mobility greatly influences the multicast gain, and in certain cases network can achieve the upper bound of multicast gain regardless of the logarithmic factor.

Multi-label learning is often applied to handle complex decision tasks, and feature selection is its essential part. The relation of labels is always ignored or not enough to consider for both multi-label learning and its feature selection. To deal with the problem, F-neighborhood rough sets are employed. Different from other methods, the original approximate space is not changed, but the relation of labels is sufficient to consider. To be specific, a multi-label decision system is discomposed into a family of single-label decision tables with the label set(first-order strategy) at first. Secondly, calculate attribute significance in the family of single-label decision tables. Third, construct an attribute significance matrix and improved attribute significance matrices to evaluate the quality of the features, then a parallel reduct is obtained with information fusion. These processes construct F-neighborhood parallel reduction algorithm for a multi-label decision system(FNPRMS). Compared with the state-of-the-arts, experimental results show that FNPRMS is effective and efficient on 9 publicly available data sets.

In proof-of-work-based (PoW-based) blockchain networks, the miners participate in a crypto-puzzle solving competition to win the reward by publishing a new block. Open mining pools attract a large number of miners for solving difficult problems together. Although the open strategy is likely to be more efficient, it makes pools susceptible to attack at the same time. In this paper, we present a game-theoretic analysis of mining pool strategy selection in order to explore the trade-off between the efficiency of openness and the vulnerability of attacks in a PoW-based blockchain network. We first model the pool mining process as a two-stage game, wherein the pools might decide whether to open or not and to attack or not. Based on the two-stage game model, we analyze the Nash equilibrium and the evolutionary stability of the mining games among pools, which uncovers the pool selection dynamics of PoW-based blockchain networks. In particular, we find that the attack behavior is the norm for a weak pool and triggers lower expected utilities when punishing the attacks more severely. Numerical simulations also support our theoretical findings as well as demonstrate the stability of the pools’ strategy selection.

Deep convolutional neural network (DCNN) based image codecs, consisting of encoder, quantizer and decoder, have achieved promising image compression results. The major challenge in learning these DCNN models lies in the joint optimization of the encoder, quantizer and decoder, as well as the adaptivity to the input images. In this paper, we proposed a DCNN architecture for image compression, where the encoder, quantizer and decoder are jointly learned. Specifically, a fully convolutional vector quantization network (VQNet) has been proposed to quantize the feature vectors of the image representation, where the representative vectors of the VQNet are jointly optimized with other network parameters through end-to-end training. While most of current DCNN-based methods were only trained once on large-scale datasets, we further perform fine-tuning of the encoder and the codes generated by the VQNet on the input images to further improve the compression performance. Extensive experimental results show that the proposed method achieves state-of-the-art compression results with simple encoder-decoder.

Immunization strategies on complex networks are effective methods to control the spreading dynamics on complex networks, which change the topology and connectivity of the underlying network, thereby affecting the dynamics process of propagation. Here, we use a non-Markovian threshold model to study the impact of immunization strategies on social contagions, in which the immune index greater than (or equal to) 0 corresponds to targeted (random) immunization, and when the immune index is less than 0, the probability of an individual being immunized is inversely related to the degree of the individual. A generalized edge-based compartmental theory is developed to analyze the dynamics of social contagions under immunization, and theoretical predictions are very consistent with simulation results. We find that increasing the immune index or increasing the immune ratio will reduce the final adoption size and increase the outbreak threshold, in other words, make the residual network after immunization not conducive to social contagions. Interestingly, enhancing the network heterogeneity is proved to help improve the immune efficiency of targeted immunization. Besides, the dependence of the outbreak threshold on the network heterogeneity is correlated with the immune ratio and immune index.

To better regulate smoking in no-smoking areas, we present a novel AI-based surveillance system for smart cities. In this paper, we intend to solve the issue of no-smoking area surveillance by introducing a framework for an AI-based smoker detection system for no-smoking areas in a smart city. Moreover, this research will provide a dataset for smoker detection problems in indoor and outdoor environments to help future research on this AI-based smoker detection system. The newly curated smoker detection image dataset consists of two classes, Smoking and NotSmoking. Further, to classify the Smoking and NotSmoking images, we have proposed a transfer learning-based solution using the pre-trained InceptionResNetV2 model. The performance of the proposed approach for predicting smokers and not-smokers was evaluated and compared with other CNN methods on different performance metrics. The proposed approach achieved an accuracy of 96.87% with 97.32% precision and 96.46% recall in predicting the Smoking and NotSmoking images on a challenging and diverse newly-created dataset. Although, we trained the proposed method on the image dataset, we believe the performance of the system will not be affected in real-time.

Despite rapid advances in the field of image denoising, heavy noise removal has remained an under-explored area. When the strength of noise becomes comparable to or even more dominating than that of signal, restoration of important structures from such heavily-contaminated images becomes more challenging. Existing model-based and learning-based image denoising techniques often cannot delivery satisfactory results, which call for new insights and algorithmic tools. In this paper, we propose to remove heavy noise with the help of similar images retrieved from datasets, which leads to a hybrid approach combining sparsity-based and learning-based methods. Specifically, the parametric sparse prior of underlying clean image is learned from the retrieved reference images and the input noisy image. Leveraging prior from the reference images, the estimated parametric model contains much more accurate information of image details. When compared against existing conventional methods, the proposed hybrid approach is more capable of restoring fine-detailed structures in the presence of heavy noise. Experimental results show that the proposed method dramatically outperforms current state-of-the-art image denoising methods both subjectively and objectively.

In terms of small objects in traffic scenes, general object detection algorithms have low detection accuracy, high model complexity, and slow detection speed. To solve the above problems, an improved algorithm (named YOLO-MXANet) is proposed in this paper. Complete-Intersection over Union (CIoU) is utilized to improve loss function for promoting the positioning accuracy of the small object. In order to reduce the complexity of the model, we present a lightweight yet powerful backbone network (named SA-MobileNeXt) that incorporates channel and spatial attention. Our approach can extract expressive features more effectively by applying the Shuffle Channel and Spatial Attention (SCSA) module into the SandGlass Block (SGBlock) module while increasing the parameters by a small number. In addition, the data enhancement method combining Mosaic and Mixup is employed to improve the robustness of the training model. The Multi-scale Feature Enhancement Fusion (MFEF) network is proposed to fuse the extracted features better. In addition, the SiLU activation function is utilized to optimize the Convolution-Batchnorm-Leaky ReLU (CBL) module and the SGBlock module to accelerate the convergence of the model. The ablation experiments on the KITTI dataset show that each improved method is effective. The improved algorithm reduces the complexity and detection speed of the model while improving the object detection accuracy. The comparative experiments on the KITTY dataset and CCTSDB dataset with other algorithms show that our algorithm also has certain advantages.

In the practical application scenarios of safety helmet detection, the lightweight algorithm You Only Look Once (YOLO) v3-tiny is easy to be deployed in embedded devices because its number of parameters is small. However, its detection accuracy is relatively low, which is why it is not suitable for detecting multi-scale safety helmets. The safety helmet detection algorithm (named SAS-YOLOv3-tiny) is proposed in this paper to balance detection accuracy and model complexity. A light Sandglass-Residual (SR) module based on depthwise separable convolution and channel attention mechanism is constructed to replace the original convolution layer, and the convolution layer of stride two is used to replace the max-pooling layer for obtaining more informative features and promoting detection performance while reducing the number of parameters and computation. Instead of two-scale feature prediction, three-scale feature prediction is used here to improve the detection effect about small objects further. In addition, an improved spatial pyramid pooling (SPP) module is added to the feature extraction network to extract local and global features with rich semantic information. Complete-Intersection over Union (CIoU) loss is also introduced in this paper to improve the loss function for promoting positioning accuracy. The results on the self-built helmet dataset show that the improved algorithm is superior to the original algorithm. Compared with the original YOLOv3-tiny, the SAS-YOLOv3-tiny has significantly improved all metrics (including Precision (P), Recall (R), Mean Average Precision (mAP), F1) at the expense of only a minor speed while keeping fewer parameters and amounts of calculation. Meanwhile, the SAS-YOLOv3-tiny algorithm shows advantages in accuracy compared with lightweight object detection algorithms, and its speed is faster than the heavyweight model.

Siamese network-based trackers consider tracking as features cross-correlation between the target template and the search region. Therefore, feature representation plays an important role for constructing a high-performance tracker. However, all existing Siamese networks extract the deep but low-resolution features of the entire patch, which is not robust enough to estimate the target bounding box accurately. In this work, to address this issue, we propose a novel high-resolution Siamese network, which connects the high-to-low resolution convolution streams in parallel as well as repeatedly exchanges the information across resolutions to maintain high-resolution representations. The resulting representation is semantically richer and spatially more precise by a simple yet effective multi-scale feature fusion strategy. Moreover, we exploit attention mechanisms to learn object-aware masks for adaptive feature refinement, and use deformable convolution to handle complex geometric transformations. This makes the target more discriminative against distractors and background. Without bells and whistles, extensive experiments on popular tracking benchmarks containing OTB100, UAV123, VOT2018 and LaSOT demonstrate that the proposed tracker achieves state-of-the-art performance and runs in real time, confirming its efficiency and effectiveness.