Search

The purpose of image dehazing is to remove the interference from weather factors in degraded images and enhance the clarity and color saturation of images to maximize the restoration of useful features. Single image dehazing is one of the most important tasks in the field of image restoration. In recent years, due to the progress of deep learning, single image dehazing has made great progress. With the success of Transformer in advanced computer vision tasks, some research studies also began to apply Transformer to image dehazing tasks and obtained surprising results. However, both the deconvolution-neural-network-based dehazing algorithm and Transformer based dehazing algorithm magnify their advantages and disadvantages separately. Therefore, this paper proposes a novel Transformer–Convolution fusion dehazing network (TCFDN), which uses Transformer's global modeling ability and convolutional neural network's local modeling ability to improve the dehazing ability. In the Transformer–Convolution fusion dehazing network, the classic self-encoder structure is used. This paper proposes a Transformer–Convolution hybrid layer, which uses an adaptive fusion strategy to make full use of the Swin-Transformer and convolutional neural network to extract and reconstruct image features. On the basis of previous research, this layer further improves the ability of the network to remove haze. A series of contrast experiments and ablation experiments not only proved that the Transformer–Convolution fusion dehazing network proposed in this paper exceeded the more advanced dehazing algorithm, but also provided solid and powerful evidence for the basic theory on which it depends. [ABSTRACT FROM AUTHOR]

To ensure the safe operation of highway traffic lines, given the imperfect feature extraction of existing road pit defect detection models and the practicability of detection equipment, this paper proposes a lightweight target detection algorithm with enhanced feature extraction based on the YOLO (You Only Look Once) algorithm. The BIFPN (Bidirectional Feature Pyramid Network) network structure is used for multi-scale feature fusion to enhance the feature extraction ability, and Varifocal Loss is used to optimize the sample imbalance problem, which improves the accuracy of road defect target detection. In the evaluation test of the model in the constructed PCD1 (Pavement Check Dataset) dataset, the mAP@.5 (mean Average Precision when IoU = 0.5) of the BV-YOLOv5S (BiFPN Varifocal Loss-YOLOv5S) model increased by 4.1%, 3%, and 0.9%, respectively, compared with the YOLOv3-tiny, YOLOv5S, and B-YOLOv5S (BiFPN-YOLOv5S; BV-YOLOv5S does not use the Improved Focal Loss function) models. Through the analysis and comparison of experimental results, it is proved that the proposed BV-YOLOv5S network model performs better and is more reliable in the detection of pavement defects and can meet the needs of road safety detection projects with high real-time and flexibility requirements. [ABSTRACT FROM AUTHOR]

This paper proposes a data valuation algorithm for inertial measurement unit-based human activity recognition (IMU-based HAR) data based on meta reinforcement learning. Unlike previous studies that received feature-level input, the algorithm in this study added a feature extraction structure to the data valuation algorithm, and it can receive raw-level inputs and achieve excellent performance. As IMU-based HAR data are multivariate time-series data, the proposed algorithm incorporates an architecture capable of extracting both local and global features by inserting a transformer encoder after the one-dimensional convolutional neural network (1D-CNN) backbone in the data value estimator. In addition, the 1D-CNN-based stacking ensemble structure, which exhibits excellent efficiency and performance on IMU-based HAR data, is used as a predictor to supervise model training. The Berg balance scale (BBS) IMU-based HAR dataset and the public datasets, UCI-HAR, WISDM, and PAMAP2, are used for performance evaluation in this study. The valuation performance of the proposed algorithm is observed to be excellent on IMU-based HAR data. The rate of discovering corrupted data is higher than 96% on all datasets. In addition, classification performance is confirmed to be improved by the suppression of discovery of low-value data. [ABSTRACT FROM AUTHOR]

Freezing of gait (FoG) is a disabling clinical phenomenon of Parkinson's disease (PD) characterized by the inability to move the feet forward despite the intention to walk. It is one of the most troublesome symptoms of PD, leading to an increased risk of falls and reduced quality of life. The combination of wearable inertial sensors and machine learning (ML) algorithms represents a feasible solution to monitor FoG in real-world scenarios. However, traditional FoG detection algorithms process all data indiscriminately without considering the context of the activity during which FoG occurs. This study aimed to develop a lightweight, context-aware algorithm that can activate FoG detection systems only under certain circumstances, thus reducing the computational burden. Several approaches were implemented, including ML and deep learning (DL) gait recognition methods, as well as a single-threshold method based on acceleration magnitude. To train and evaluate the context algorithms, data from a single inertial sensor were extracted using three different datasets encompassing a total of eighty-one PD patients. Sensitivity and specificity for gait recognition ranged from 0.95 to 0.96 and 0.80 to 0.93, respectively, with the one-dimensional convolutional neural network providing the best results. The threshold approach performed better than ML- and DL-based methods when evaluating the effect of context awareness on FoG detection performance. Overall, context algorithms allow for discarding more than 55% of non-FoG data and less than 4% of FoG episodes. The results indicate that a context classifier can reduce the computational burden of FoG detection algorithms without significantly affecting the FoG detection rate. Thus, implementation of context awareness can present an energy-efficient solution for long-term FoG monitoring in ambulatory and free-living settings. [ABSTRACT FROM AUTHOR]

Fault diagnosis methods based on deep learning and big data have achieved good results on rotating machinery. However, the conventional deep learning method of bearing fault diagnosis is mostly based on laboratory artificial simulation data, and there is an error with actual fault data, which will reduce the generalization performance of the deep learning method. In addition, labeled data are very precious in real industrial environment. Due to expensive equipment and personnel safety issues, it is difficult to obtain a large amount of high-quality fault labeling data. Therefore, in this paper, we propose a metric-based meta-learning method named Reinforce Relation Network (RRN) for diagnosing bearing faults with few-shot samples. In the proposed method, a 1D convolution neural network is used to extract fault features, and a metric learner is used to predict the similarity between samples under different transfer conditions. Label smoothing and the Adabound algorithm are utilized to further improve the performance of network classification. The performance of the proposed method is verified on a dataset which contains artificial damage and natural damage data. The comparison studies with other methods demonstrate the superiority of the proposed method in the few-shot scenario. [ABSTRACT FROM AUTHOR]

Transfer learning can enhance classification performance of a target domain with insufficient training data by utilizing knowledge relating to the target domain from source domain. Nowadays, it is common to see two or more source domains available for knowledge transfer, which can improve performance of learning tasks in the target domain. However, the classification performance of the target domain decreases due to mismatching of probability distribution. Recent studies have shown that deep learning can build deep structures by extracting more effective features to resist the mismatching. In this paper, we propose a new multi-source deep transfer neural network algorithm, MultiDTNN, based on convolutional neural network and multi-source transfer learning. In MultiDTNN, joint probability distribution adaptation (JPDA) is used for reducing the mismatching between source and target domains to enhance features transferability of the source domain in deep neural networks. Then, the convolutional neural network is trained by utilizing the datasets of each source and target domain to obtain a set of classifiers. Finally, the designed selection strategy selects classifier with the smallest classification error on the target domain from the set to assemble the MultiDTNN framework. The effectiveness of the proposed MultiDTNN is verified by comparing it with other state-of-the-art deep transfer learning on three datasets. [ABSTRACT FROM AUTHOR]

In this study, we propose a new intelligent system to automatically quantify the morphological parameters of the lamina cribrosa (LC) of the optical coherence tomography (OCT), including depth, curve depth, and curve index from OCT images. The proposed system consisted of a two-stage deep learning (DL) model, which was composed of the detection and the segmentation models as well as a quantification process with a post-processing scheme. The models were used to solve the class imbalance problem and obtain Bruch's membrane opening (BMO) as well as anterior LC information. The detection model was implemented by using YOLOv3 to acquire the BMO and LC position information. The Attention U-Net segmentation model is used to compute accurate locations of the BMO and LC curve information. In addition, post-processing is applied using polynomial regression to attain the anterior LC curve boundary information. Finally, the numerical values of morphological parameters are quantified from BMO and LC curve information using an image processing algorithm. The average precision values in the detection performances of BMO and LC information were 99.92% and 99.18%, respectively, which is very accurate. A highly correlated performance of R2 = 0.96 between the predicted and ground-truth values was obtained, which was very close to 1 and satisfied the quantification results. The proposed system was performed accurately by fully automatic quantification of BMO and LC morphological parameters using a DL model. [ABSTRACT FROM AUTHOR]

Recently, deep learning approaches, especially convolutional neural networks (CNNs), have attracted extensive attention in iris recognition. Though CNN-based approaches realize automatic feature extraction and achieve outstanding performance, they usually require more training samples and higher computational complexity than the classic methods. This work focuses on training a novel condensed 2-channel (2-ch) CNN with few training samples for efficient and accurate iris identification and verification. A multi-branch CNN with three well-designed online augmentation schemes and radial attention layers is first proposed as a high-performance basic iris classifier. Then, both branch pruning and channel pruning are achieved by analyzing the weight distribution of the model. Finally, fast finetuning is optionally applied, which can significantly improve the performance of the pruned CNN while alleviating the computational burden. In addition, we further investigate the encoding ability of 2-ch CNN and propose an efficient iris recognition scheme suitable for large database application scenarios. Moreover, the gradient-based analysis results indicate that the proposed algorithm is robust to various image contaminations. We comprehensively evaluated our algorithm on three publicly available iris databases for which the results proved satisfactory for real-time iris recognition. [ABSTRACT FROM AUTHOR]