JournalTOCs

The Visual Computer
Journal Prestige (SJR): 0.401

Citation Impact (citeScore): 2
Number of Followers: 3

Hybrid journal (It can contain Open Access articles)
ISSN (Print) 1432-2315 - ISSN (Online) 0178-2789
Published by Springer-Verlag

[2467 journals]

Automatic colorization for Thangka sketch-based paintings
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Thangka is a kind of ancient painting that originated from Tibetan culture. Considering its value in cultural and historical fields, digital protection and restoration of Thangka images are important. In this paper, we propose an automatic colorization framework for a Thangka sketch. First, we propose a network structure with a feedback generator to restore the true color of a Thangka painting based on semantic matching. Second, we design a color gradient algorithm for Thangka coloring. Finally, we construct an automatic coloring framework for Thangka paintings that supports user customization. The experimental results show that the algorithm has a highly accurate response to a user’s selection process. Compared with traditional methods, the proposed method improves the image quality by 22.3% on average. The data for our approach are publicly available at https://github.com/wangfubo123/SMAC-CGAN.
  PubDate: 2023-03-21
Learning local contextual features for 3D point clouds semantic
segmentation by attentive kernel convolution
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Unlike 2D images that are represented in regular grids, 3D point clouds are irregular and unordered, hence directly applying convolution neural networks (CNNs) to process point clouds is quite challenging. In this paper, we propose a novel deep neural network named AKNet to achieve point cloud semantic segmentation. The key to our AKNet is the attentive kernel convolution (AKConv), which is a deformed convolution operation for perceiving sufficient local context of 3D scenes. AKConv first constructs the Basic Weight Units that are robust to point’s ordering. Then, for capturing the more distinctive local features, the convolution kernels of AKConv are associated with Attentive Weight Units through the self-attentive function acting on Basic Weight Units. Furthermore, 3D point clouds provide richer geometric shape information, which is helpful to recognize objects. However, inputting only raw point features to the convolution function could cause geometric information loss. Thus, we utilize augmented features as input of AKConv. Besides, to preserve the semantic information from the encoding to decoding layers, we introduce the backward encoding (BE) mechanism by utilizing higher-layer semantic features. We conduct experiments on three large-scale point clouds datasets. The experimental results demonstrate that our AKNet outperforms state-of-the-art (SOTA) networks.
  PubDate: 2023-03-19
ODSPC: deep learning-based 3D object detection using semantic point cloud
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Three-dimensional object detection plays a key role in autonomous driving, which becomes extremely challenging in occlusion situations. This paper presents a novel multimodal 3D object detection framework which fuses visual semantic information and depth point cloud information to accurately detect targets with distant object features and occlusion situations. The framework consists of the four steps. Firstly, an improved semantic segmentation network is used to extract semantic information of objects containing similar features. Secondly, semantic images and point clouds are combined to generate pixel-level fusion data so that the semantic information and training capability of sparse and far-point clouds can be improved. Thirdly, a deep learning-based point cloud classification network is used for training of the fused data to output accurate detection frames. Fourthly, an extended Kalman filter is incorporated into point cloud prediction for image-based object detection to further enhance the robustness of object detection. Both Cityscapes and KITTI datasets are used in ablation study and experiments to validate the effectiveness of the proposed framework.
  PubDate: 2023-03-18
YOLOF-F: you only look one-level feature fusion for traffic sign detection
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract This paper proposes a detector that focuses on multi-scale detection problems and effectively enhances the detection performance to solve the problem that is hard to detect minor traffic signs. This detector, called YOLOF-F (you only look one-level feature fusion), is a single-stage detector that extracts multi-scale feature information from a single layer of fusion feature. First, we propose FFM (feature fusion module) to fuse different scales. Next, we offer a new encoder CDE (corner dilated encoder) to enhance the angular point information in the feature map, improve position regression accuracy, and maintain a faster detection speed. Finally, YOLOF-F achieved 74.57% and 77.23% of the AP on the GTSDB and CTSD datasets and reached 32 FPS. Extensive experiments validate that YOLOF-F is faster and more effective than most traffic sign detection methods.
  PubDate: 2023-03-17
ODRP: a new approach for spatial street sign detection from EXIF using
deep learning-based object detection, distance estimation, rotation and
projection system
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Geographical information systems (GIS) are the systems where spatial data are stored and analyzed. The most important raw material in GIS is spatial data. Thus, it is essential to collect and update these data. On the other hand, exchangeable image file (EXIF) format is a special file format that contains camera direction, date-time information and GPS location provided by a digital camera that captures the images. Transferring the objects in EXIF data sets with absolute coordinates on the earth significantly contributes to GIS. In this study, a new hybrid approach, ODPR, which utilizes object detection (O), distance estimation (D), rotation (R) and projection (P) methods, is proposed to detect street sign objects in EXIF with their locations. The performance of the proposed approach has been examined on the natural EXIF data sets obtained from the Kayseri Metropolitan Municipality. In the proposed approach, a deep learning method detects a street sign object in the EXIF. Then, the object’s distance is calculated at the point where the photograph is taken. Finally, the spatial location of the detected object on the earth is calculated using distance, direction and GPS data with rotation and projection methods. In the proposed ODRP approach, the performances of convolutional neural network (CNN)-based Faster R-CNN, YOLO V5, YOLO V6 and transformer-based DETR models as deep learning models for object detection are examined. The F1 score metric is widely used to examine the performance of methods in deep learning models. The performances of the proposed approaches are reviewed according to the F1 score values, and ODRP Faster R-CNN, YOLO V5, YOLO V6 and DETR approaches achieved F1 scores of 0.909, 0.956, 0.948 and 0.922, respectively. In addition, to overcome the variability of light and background mixing problems, an improved supervised learning method (ISL) is proposed. Thanks to ISL, ODRP Faster R-CNN, YOLO V5, YOLO V6, and DETR approaches have reached 0.965, 0.985, 0.969 and 0.942 f1 scores, respectively. The proposed ODRP Faster R-CNN, YOLO V5, YOLO V6 and DETR approaches found the location of the street sign object to be 11434.76, 12818.39, 12454.63 and 9843.57 ms closer to its position on earth than the classical method, which considers the location of the EXIF, respectively. Regarding time cost, the ODRP Faster R-CNN, YOLO V5, YOLO V6 and DETR analyze EXIF data at an average of 0.99, 0.42, 0.41 and 0.53 s, respectively. The run time of the ODRP YOLO V5 and V6 approaches is almost equal to each other, and it works approximately 2.5 times faster than the ODRP Faster R-CNN method. Consequently, ODRP YOLO V5 outperforms ODRP Faster R-CNN, YOLO V6 and DETR for detecting the spatial location of street sign objects in EXIF and the F1 score.
  PubDate: 2023-03-17
Physical flexibility detection under complex backgrounds using ED-Former
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Anteflexion angle is an essential health indicator to assess physical flexibility, and image recognition has become an effective method to detect physical anteflexion angle. In this paper, we present a physical flexibility detection algorithm under complex backgrounds, based on edge detection with transformers (ED-Former). The algorithm adopts the lightweight self-attention mechanism from the transformer architecture and combines it with an edge detection network. Additionally, we design the angle protector, which is added to the output to reduce the effect of weak edges in the problem image. Then, based on the obtained edge information, we propose a low-complexity ergonomics-based physical regions combination strategy to calculate the anteflexion angle. We select the regions of the shoulder, waist, and knee, and calculate the angle based on the relation between these regions. The experimental results show that the ED-Former has an excellent performance in physical edge detection tasks—notably producing an F1 score of 0.833 and outperforming state-of-the-art CNNs and transformers on these tasks. Furthermore, our network is effective in preventing adversarial attacks. Overall, we find that the average error of anteflexion angle with respect to the true is 1.267°, well within the 5° requirement for physical flexibility detection.
  PubDate: 2023-03-17
Query semantic reconstruction for background in few-shot segmentation
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Few-shot segmentation (FSS) aims to segment unseen classes using a few annotated samples. Typically, a prototype representing the foreground class is extracted from annotated support image(s) and is matched to features representing each pixel in the query image. However, models learnt in this way are insufficiently discriminatory, and often produce false positives: misclassifying background pixels as foreground. Some FSS methods try to address this issue by using the background in the support image(s) to help identify the background in the query image. However, the backgrounds of these images are often quite distinct, and hence, the support image background information is uninformative. This article proposes a method, QSR, that extracts the background from the query image itself, and as a result is better able to discriminate between foreground and background features in the query image. This is achieved by modifying the training process to associate prototypes with class labels including known classes from the training data and latent classes representing unknown background objects. This class information is then used to extract a background prototype from the query image. To successfully associate prototypes with class labels and extract a background prototype that is capable of predicting a mask for the background regions of the image, the machinery for extracting and using foreground prototypes is induced to become more discriminative between different classes. Experiments achieves state-of-the-art results for both 1-shot and 5-shot FSS on the PASCAL- \(5^{i}\) and COCO- \(20^{i}\) dataset. As QSR operates only during training, results are produced with no extra computational complexity during testing.
  PubDate: 2023-03-16
A method of smoothing laser spot deformation
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Spot size is an important parameter in laser-quality detection. However, the size of the spot is easily affected by environmental factors such as light. In this paper, we propose improved Kalman filter algorithms (IKF) with prior information to reduce the negative influence of noise on the spot size. At the same time, the filtering process and the method of determining the initial filter value are given. Finally, the experimental results on the synthetic dataset and real-world dataset confirm that the proposed IKF performs better than the compared methods.
  PubDate: 2023-03-15
Vehicle object counting network based on feature pyramid split attention
mechanism
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract In recent years, real-time vehicle congestion detection has become a hot research topic in the field of transportation due to the frequent occurrence of highway traffic jams. Vehicle congestion detection generally adopts a vehicle counting algorithm based on object detection, but it is not effective in scenarios with large changes in vehicle scale, dense vehicles, background clutter, and severe occlusion. A vehicle object counting network based on a feature pyramid split attention mechanism is proposed for accurate vehicle counting and the generation of high-quality vehicle density maps in highly congested scenarios. The network extracts rich contextual features by using blocks at different scales, and then obtains a multi-scale feature mapping in the channel direction using kernel convolution of different sizes, and uses the channel attention module at different scales separately to allow the network to focus on features at different scales to obtain an attention vector in the channel direction to reduce mis-estimation of background information. Experiments on the vehicle datasets TRANCOS, CARPK, and HS-Vehicle show that the proposed method outperforms most existing counting methods based on detection or density estimation. The relative improvement in MAE metrics is 90.5% for the CARPK dataset compared to Fast R-CNN and 73.0% for the HS-Vehicle dataset compared to CSRNet. In addition, the method is also extended to count other objects, such as pedestrians in the ShanghaiTech dataset, and the proposed method effectively reduces the misrecognition rate and achieves higher counting performance compared to the state-of-the-art methods.
  PubDate: 2023-03-14
TFTSVM: near color recognition of polishing red lead via SVM based on
threshold and feature transform
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract With the extensive application of machine vision in the manufacturing industry, target region recognition in complex industrial scenes is becoming a vital research territory. In the automatic polishing of molds, polishing red lead, as an auxiliary tool for polishing positioning, can intuitively determine the areas to be polished. Its bright color information are very suitable for vision-based recognition. Due to the interference of the near color in the polishing environment, the traditional color recognition method has the appearance of over-segmentation. In this paper, we propose a novel near-color recognition method via SVM based on threshold and feature transform (TFTSVM) to improve the identification accuracy of polishing red lead. Specifically, this method adopts a threshold-based color recognition algorithm to extract two kinds of color features of red lead color and its near color in HSV color space and skillfully finds it is distinguishable in three dimensions. To reduce the computational complexity, a machine learning segmentation model is constructed, which realizes dimension reduction by integrating the feature transformation method of sample transformation and projection transformation to achieve the best segmentation effect. Experimental results on self-established dataset demonstrate that the proposed method has an excellent identification effect on the red lead area in the field polishing environment and also shows good robustness under the condition that there are reflections on the mold surface. It meets the requirements of mechanical arm polishing and improves the safety and reliability of automatic polishing. In addition, we also compare different machine learning algorithms and advanced studies to verify the correctness of the algorithm. This method also provides a reference for realizing near-color recognition in complex industrial environments.
  PubDate: 2023-03-14
Visual image encryption scheme based on inter-intra-block scrambling and
weighted diffusion
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract This paper presents an image encryption scheme with data and appearance security, by adopting inter-intra-block scrambling and weighted diffusion. The 2D robust hyper-chaotic map with flexible geometric distribution and rich hyper-chaotic parameter space is employed to generate the key stream for encryption, by considering the characteristics of plaintext. The plain image is first preprocessed by Huffman coding for getting compressed image. Then, the compressed image is divided into four sub-blocks and is further permuted by the designed inter-intra-block scrambling scheme, which can improve the scrambling effect by making the pixel far away from the original adjacent pixels. After that, a weighted diffusion method strongly related to plaintext and key stream is introduced to diffuse the shuffled image to obtain the noise-like cipher image. And in pursuit of higher security, the meaningless noise-like image is embedded into host image to create the visually meaningful cipher image. A series of experiment tests and analyses are carried out to further demonstrate the excellent performances of the encryption scheme.
  PubDate: 2023-03-11
ImGeo-VoteNet: image and geometry co-supported VoteNet for RGB-D object
detection
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Depth cameras are becoming affordable and widely used to capture depth information in various real-world scenarios. However, depth-based object detectors rarely fully explore the geometry of objects, as well as simply concatenate textures and depths. By fully exploiting images and geometries of objects, we propose a VoteNet-based RGB-D object detection neural network (call ImGeo-VoteNet for short) to address the adverse situation of occluded and similar objects in indoor scenes. First, image votes are generated based on a set of candidate boxes from 2D detectors in RGB images to support the subsequent voting. Second, we transform the depths of any input scene to a point cloud representation for better using its geometry information. Third, we design three modules to capture multi-level contextual information at the point level, the object level and the global scene level, respectively, for alleviating data loss and distinguishing similar objects. Extensive experiments on the benchmark dataset show that ImGeo-VoteNet obtains a better accuracy of 3D object detection in complex indoor scenes than the state-of-the-art methods. For example, ImGeo-VoteNet achieves the improvement of +5.8 mAP over VoteNet. Source code will be available upon publication.
  PubDate: 2023-03-10
Embedded 3D reconstruction of dynamic objects in real time for maritime
situational awareness pictures
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Assessing the security status of maritime infrastructures is a key factor for maritime safety and security. Facilities such as ports and harbors are highly active traffic zones with many different agents and infrastructures present, like containers, trucks or vessels. Conveying security-related information in a concise and easily understandable format can support the decision-making process of stakeholders, such as port authorities, law enforcement agencies and emergency services. In this work, we propose a novel real-time 3D reconstruction framework for enhancing maritime situational awareness pictures by joining temporal 2D video data into a single consistent display. We introduce and verify a pipeline prototype for dynamic 3D reconstruction of maritime objects using a static observer and stereoscopic cameras on an GPU-accelerated embedded device. A simulated dataset of a harbor basin was created and used for real-time processing. Usage of a simulated setup allowed verification against synthetic ground-truth data. The presented pipeline runs entirely on a remote, low-power embedded system with \(\sim \) 6 Hz. A Nvidia Jetson Xavier AGX module was used, featuring 512 CUDA-cores, 16 GB memory and an ARMv8 64-bit octa-core CPU.
  PubDate: 2023-03-10
Monocular visual-inertial odometry leveraging point-line features with
structural constraints
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Structural geometry constraints, such as perpendicularity, parallelism and coplanarity, are widely existing in man-made scene, especially in Manhattan scene. By fully exploiting these structural properties, we propose a monocular visual-inertial odometry (VIO) using point and line features with structural constraints. First, a coarse-to-fine vanishing points estimation method with line segment consistency verification is presented to classify lines into structural and non-structural lines accurately with less computation cost. Then, to get precise estimation of camera pose and the position of 3D landmarks, a cost function which combines structural line constraints with feature reprojection residual and inertial measurement unit residual is minimized under a sliding window framework. For geometric representation of lines, Plücker coordinates and orthonormal representation are utilized for 3D line transformation and non-linear optimization respectively. Sufficient evaluations are conducted using two public datasets to verify that the proposed system can effectively enhance the localization accuracy and robustness than other existing state-of-the-art VIO systems with acceptable time consumption.
  PubDate: 2023-03-10
An adaptive loss weighting multi-task network with attention-guide
proposal generation for small size defect inspection
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Computer vision-based detection approaches have been widely used in defect inspection tasks. However, identifying small-sized defects is still a challenge for most existing methods. It is mainly because: (1) the existing methods fail to extract sufficient information from the small-sized defects; (2) the existing detectors cannot generate effective region proposals for small-sized defects, which results in a low recall rate. To address the above issues, an adaptive loss weighting multi-task model with attention-guide proposal generation is proposed. First, the proposed multi-task model can excavate contextual information to enrich the feature information of small-sized defect areas, enhancing the model’s representation capability. Additionally, to improve the recall rate of small-sized defects, an object attention-guide proposal generation module is proposed by leveraging object attention to guide the confidence enhancement of small-sized defects, which can generate more high-quality region proposals for small-sized defects. Finally, to speed up the joint optimization of the proposed multi-task framework, an adaptive loss weighting algorithm is proposed to learn the optimal combination of multi-task loss functions by maintaining the gradient direction consistency and tuning each task’s loss magnitude. The experimental results on the two public defect datasets demonstrate that the proposed method outperforms other state-of-the-art methods.
  PubDate: 2023-03-09
Ensemble graph Laplacian-based anomaly detector for hyperspectral imagery
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Hyperspectral anomaly detection is an alluring topic in hyperspectral image processing. As one of the most famous hyperspectral anomaly detection algorithms, Reed-Xiaoli detector is widely studied since it is understandable and easy to implement. However, the estimation of inverse covariance matrix may be time-consuming and easily corrupted by the anomalies. To solve these problems, we propose a novel ensemble graph Laplacian-based anomaly detector which comprises two main steps. Firstly, a multiple random sampling strategy is applied to improve the detection accuracies and robustness. Secondly, we can obtain multiple detection results through a graph Laplacian-based solution, and these results are further fused through ensemble learning. Experimental results on one simulated and two real hyperspectral datasets demonstrate the superiority of the proposed method.
  PubDate: 2023-03-08
GADA-SegNet: gated attentive domain adaptation network for semantic
segmentation of LiDAR point clouds
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract We propose GADA-SegNet, a gated attentive domain adaptation network for semantic segmentation of LiDAR point clouds. Unlike most of existing methods that learn fully from point-wise annotations, our GADA-SegNet attempts to learn from labeled data first and then transfer itself smoothly to unlabeled data. We have three key contributions to bridge the domain gap between the labeled data and the unlabeled yet unseen data. First, we design a new gated connection module that can filter out noise and domain-private features from the low-level features, for better high- and low-level feature fusion. Second, we introduce a multi-scale attention module that can ease the large-scale variation of objects and class imbalance in complex scenes to reduce the class-level domain gap. Third, we develop a shared domain discriminator to implement the class-level domain discrimination for large-scale LiDAR point clouds. Experiments on both synthetic-to-real and real-to-real scenarios show clear improvements of our GADA-SegNet over its competitors.
  PubDate: 2023-03-08
Multi-stage adaptive rank statistic pruning for lightweight human 3D mesh
recovery model
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract We present a rank statistic adaptive multi-stage pruning method to find lightweight neural networks for 3D human mesh recovery while minimizing accuracy drop. We observe that some feature maps often have prominent low-rank patterns regardless of input human images. Furthermore, even after pruning, feature channels that should have been pruned according to pruning criteria frequently re-appear in test time. From these observations, we design rank statistic adaptive multi-stage pruning; thereby, we can prune more filters with recovering mesh reconstruction accuracy. We demonstrate that, for DenseNet-121, 60.0% of parameters and 67.9% of FLOPs are saved while maintaining comparable accuracy to that of the original full model. This is a notable improvement compared to the competing method based on the L1 filter pruning, where the error is increased by 17.55% at the same pruning rate.
  PubDate: 2023-03-08
Hybrid dilated multilayer faster RCNN for object detection
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract Faster region-based convolution neural network (Faster RCNN) architecture was proposed as an efficient object detection method, wherein a CNN is used to extract image features. However, CNNs require a large number of learning parameters, and an excessive amount of pooling layers lead to a loss of information on small objects, which may affect efficiency. In this study, we proposed a hybrid dilated multilayer Faster RCNN model to address this problem. The key contributions of this work are summarized as follows: (1) We substituted a hybrid dilated CNN (HDC) model for the VGG16 network used in the original Faster RCNN architecture to extract features and ensure portability. We also used a LeakyReLU activation function to improve the mapping ability of negative input information to detect objects rapidly and accurately. (2) We used a multilayer feature spatial pyramid to convert single-scale features into multi-scale features, and higher-resolution information was obtained through a deconvolutional network to achieve more accurate object detection. (3) We conducted experiments to verify the performance of the proposed HDMF-RCNN model using the Microsoft COCO data set. The results indicated that the accuracy of HDMF-RCNN was 8.12% greater than that of the traditional Faster RCNN model, and the training loss and training time were lower by 44.64% and 39.46% on average, respectively. Overall, the results verified that HDMF-RCNN can significantly improve on the efficiency of existing object detection methods. As an independent feature extraction network, HDC can be adapted to different network frameworks.
  PubDate: 2023-03-07
Wire rope defect identification based on ISCM-LBP and GLCM features
- Free pre-print version: Loading...
  Rate this result: What is this?
  
  Please help us test our new pre-print finding feature by giving the pre-print link a rating.
  A 5 star rating indicates the linked pre-print has the exact same content as the published article.
  
  Abstract: Abstract The traditional local binary pattern (LBP) is susceptible to the influence of the centre pixel and noise and cannot accurately identify wire rope surface defects. To solve this problem, an image segmentation-based central multiscale local binary pattern (ISCM-LBP) and grey level cooccurrence matrix (GLCM) feature fusion method is proposed in this paper for defect recognition. Image segmentation and multiple scales are introduced into the local binary pattern algorithm to improve the image detail description and suppress noise sensitivity. Second, the centre pixel is connected with the neighbourhood pixel to enhance the robustness of the centre pixel. To further improve the image integrity description, PCA dimensionality reduction and GLCM feature fusion are performed on the features extracted by the ISCM-LBP algorithm, and the steel wire rope surface defects are identified by a support vector machine classifier. Experimental results show that the overall recognition rate reaches 97.5%, which is at least 5% higher than that of other algorithms and can effectively identify various defects on the surface of wire rope.
  PubDate: 2023-03-06