Standard Convolution Research Articles

Physics-Inspired Adaptions to Low-Parameter Neural Network Weather Forecast Systems

Abstract Recently, there has been a surge of research on data-driven weather forecasting systems, especially applications based on convolutional neural networks (CNNs). These are usually trained on atmospheric data represented on regular latitude–longitude grids, neglecting the curvature of Earth. We assess the benefit of replacing the standard convolution operations with an adapted convolution operation that takes into account the geometry of the underlying data (SphereNet convolution), specifically near the poles. Additionally, we assess the effect of including the information that the two hemispheres of Earth have “flipped” properties—for example, cyclones circulating in opposite directions—into the structure of the network. Both approaches are examples of physics-informed machine learning. The methods are tested on the WeatherBench dataset, at a resolution of ∼1.4°, which is higher than many previous studies on CNNs for weather forecasting. For most lead times up to day +10 for 500-hPa geopotential and 850-hPa temperature, we find that using SphereNet convolution or including hemisphere-specific information individually leads to improvement in forecast skill. Combining the two methods typically gives the highest forecast skill. Our version of SphereNet is implemented flexibly and scales well to high-resolution datasets but is still significantly more expensive than a standard convolution operation. Finally, we analyze cases with high forecast error. These occur mainly in winter and are relatively consistent across different training realizations of the networks, pointing to flow-dependent atmospheric predictability.

Automatic Identification and Tracking Method of Case-Related Vehicles Based on Computer Vision Algorithm

Abstract Amidst the rapid advancement of artificial intelligence, computer vision algorithms have found extensive applications across various societal sectors. This paper presents the development of an automatic vehicle identification algorithm for crime-related scenarios, leveraging enhancements in computer vision technology. Initially, we refine the multi-scale feature fusion within the YOLOv4 architecture, subsequently substituting the standard convolution in the feature extraction network with depth-separable convolution to minimize parameter computation. Furthermore, we replace the conventional CIOU target localization loss function with EIOU to expedite model convergence. To address the issue of target vehicle detection failures, we incorporate the Kalman filter algorithm, ensuring precise tracking. Our experimental analysis, which utilizes both target detection and multi-target tracking evaluation indices, demonstrates that the modified YOLOv4 algorithm excels in recall, precision, and average IOU metrics compared to other algorithms, with a mean Average Precision (mAP) of 95.68% and an average detection speed of 0.039 seconds per image, satisfying real-time operational criteria. Vehicle tracking efficacy, compared using the YOLOv4-based detector before and after the modifications, shows significant improvements, indicated by a reduction in Identification Switches (IDS) across all video samples. This study introduces a robust method for the accurate identification and localization of vehicles involved in criminal activities, significantly enhancing case resolution efficiency.

Block-Wise Separable Convolutions: An Alternative Way to Factorize Standard Convolutions

Block-Wise Separable Convolutions: An Alternative Way to Factorize Standard Convolutions

Adaptive Blind Super-Resolution Network for Spatial-Specific and Spatial-Agnostic Degradations.

Prior methodologies have disregarded the diversities among distinct degradation types during image reconstruction, employing a uniform network model to handle multiple deteriorations. Nevertheless, we discover that prevalent degradation modalities, including sampling, blurring, and noise, can be roughly categorized into two classes. We classify the first class as spatial-agnostic dominant degradations, less affected by regional changes in image space, such as downsampling and noise degradation. The second class degradation type is intimately associated with the spatial position of the image, such as blurring, and we identify them as spatial-specific dominant degradations. We introduce a dynamic filter network integrating global and local branches to address these two degradation types. This network can greatly alleviate the practical degradation problem. Specifically, the global dynamic filtering layer can perceive the spatial-agnostic dominant degradation in different images by applying weights generated by the attention mechanism to multiple parallel standard convolution kernels, enhancing the network's representation ability. Meanwhile, the local dynamic filtering layer converts feature maps of the image into a spatially specific dynamic filtering operator, which performs spatially specific convolution operations on the image features to handle spatial-specific dominant degradations. By effectively integrating both global and local dynamic filtering operators, our proposed method outperforms state-of-the-art blind super-resolution algorithms in both synthetic and real image datasets.

A New Siamese Heterogeneous Convolutional Neural Networks Based on Attention Mechanism and Feature Pyramid.

Accuracy and speed are the most important indexes for evaluating many object tracking algorithms. However, when constructing a deep fully convolutional neural network (CNN), the use of deep network feature tracking will cause tracking drift due to the effects of convolution padding, receptive field (RF), and overall network step size. The speed of the tracker will also decrease. This article proposes a fully convolutional siamese network object tracking algorithm that combines the attention mechanism with the feature pyramid network (FPN), and uses heterogeneous convolution kernels to reduce the amount of calculations (FLOPs) and parameters. The tracker first uses a new fully CNN to extract image features, and introduces a channel attention mechanism in the feature extraction process to improve the representation ability of convolutional features. Then use the FPN to fuse the convolutional features of high and low layers, learn the similarity of the fused features, and train the fully CNNs. Finally, the heterogeneous convolutional kernel is used to replace the standard convolution kernel to improve the speed of the algorithm, thereby making up for the efficiency loss caused by the feature pyramid model. In this article, the tracker is experimentally verified and analyzed on the VOT-2017, VOT-2018, OTB-2013, and OTB-2015 datasets. The results show that our tracker has achieved better results than the state-of-the-art trackers.

A YOLOv8-Based Approach for Real-Time Lithium-Ion Battery Electrode Defect Detection with High Accuracy

Targeting the issue that the traditional target detection method has a high missing rate of minor target defects in the lithium battery electrode defect detection, this paper proposes an improved and optimized battery electrode defect detection model based on YOLOv8. Firstly, the lightweight GhostCony is used to replace the standard convolution, and the GhostC2f module is designed to replace part of the C2f, which reduces model computation and improves feature expression performance. Then, the coordinate attention (CA) module is incorporated into the neck network, amplifying the feature extraction efficiency of the improved model. Finally, the EIoU loss function is employed to swap out the initial YOLOv8 loss function, which improves the regression performance of the network. The empirical findings demonstrate that the enhanced model exhibits increments in crucial performance metrics relative to the original model: the precision rate is elevated by 2.4%, the recall rate by 2.3%, and the mean average precision (mAP) by 1.4%. The enhanced model demonstrates a marked enhancement in the frames per second (FPS) detection rate, significantly outperforming other comparative models. This evidence indicates that the enhanced model aligns well with the requirements of industrial development, demonstrating substantial practical value in industrial applications.

DAUnet: A U-shaped network combining deep supervision and attention for brain tumor segmentation

In MRI images, the brain tumor area varies greatly between individuals, and only relying on the judgment of clinicians is prone to misdiagnosis and misjudgment. Consequently, utilizing computer-aided diagnosis is of utmost significance in assisting clinicians with delineating the tumor region. Brain tumor MRI images are 3D images, and traditional segmentation methods tend to lose key information. Therefore, this paper proposes DAUnet, an U-shaped network for brain tumor MRI image segmentation combining deep supervision and convolutional attention. First, a module consisting of a Bottleneck module and attention (BA) module is designed. Here the attention not only uses spatial and channel (SC) attention but also adds residual connection, which is called 3D SC attention. Second, to enlarge the feature map receptive field without changing its resolution, a module consists of standard convolution and atrous spatial pyramid (CASP) module is designed. The feature map information is adjusted by standard convolution, subsequently, the feature map is provided as input to the ASP module. The CASP module fuses the features extracted by downsampling and performs upsampling operation, which strengthens the correlation between different layers of the network. Finally, using deep supervision as an auxiliary branch of the U-shaped network, it combines deep learning and regularization techniques to supervise the model during training, automatically finer parameters, and make the model fit better. Through experiments on BraTS 2020 and FeTS 2021 and comparison with other advanced methods, it has been demonstrated that DAUnet achieves precise segmentation of tumor regions in brain MRI images.

A lightweight YOLOv7 insulator defect detection algorithm based on DSC-SE.

As the UAV(Unmanned Aerial Vehicle) carrying target detection algorithm in transmission line insulator inspection, we propose a lightweight YOLOv7 insulator defect detection algorithm for the problems of inferior insulator defect detection speed and high model complexity. Firstly, a lightweight DSC-SE module is designed using a DSC(Depthwise Separable Convolution) fused SE channel attention mechanism to substitute the SC(Standard Convolution) of the YOLOv7 backbone extraction network to decrease the number of parameters in the network as well as to strengthen the shallow network's ability to obtain information about target features. Then, in the feature fusion part, GSConv(Grid Sensitive Convolution) is used instead of standard convolution to further lessen the number of parameters and the computational effort of the network. EIoU-loss(Efficient-IoU) is performed in the prediction head part to make the model converge faster. According to the experimental results, the recognition accuracy rate of the improved model is 95.2%, with a model size of 7.9M. Compared with YOLOv7, the GFLOPs are reduced by 54.5%, the model size is compressed by 37.8%, and the accuracy is improved by 4.9%. The single image detection time on the Jetson Nano is 105ms and the capture rate is 13FPS. With guaranteed accuracy and detection speed, it meets the demands of real-time detection.

On the fractional space-time Fourier transforms

ABSTRACT The notion of a fractional space-time Fourier transform (FSFT) is outlined in this paper, and the properties of invertibility, linearity, Plancherel and others are derived. By establishing the relationship between the FSFT and space-time Fourier transform, a directional uncertainty principle (UP) and its specialization to coordinates are proved. Moreover, the Hardy UP of the FSFT is also obtained. The fractional Mustard convolution for space-time valued signals is introduced and written in terms of the standard convolution. Finally, the FSFT is employed in solving a partial differential equation in space-time analysis.

Remainder terms of a nonlocal Sobolev inequality

AbstractIn this note, we study a nonlocal version of the Sobolev inequality where is the best constant, denotes the standard convolution and denotes the classical Sobolev space with respect to the norm . By using the nondegeneracy property of the extremal functions, we prove that the existence of the gradient type remainder term and a reminder term in the weak ‐norm of above inequality for all with .

Detection of the Grassland Weed Phlomoides umbrosa Using Multi-Source Imagery and an Improved YOLOv8 Network

Grasslands are the mainstay of terrestrial ecosystems and crucial ecological barriers, serving as the foundation for the development of grassland husbandry. However, the frequent occurrence of poisonous plants in grasslands weakens the stability of grassland ecosystems and constrains the growth of grassland livestock husbandry. To achieve early detection of the grassland weed Phlomoides umbrosa (Turcz.) Kamelin & Makhm, this study improves the YOLO-v8 model and proposes a BSS-YOLOv8 network model using UAV images. Using UAV, we can obtain early-stage image data of P. umbrosa and build a seedling dataset. To address challenges such as the complex grassland background and the dwarf seedlings of P. umbrosa, this study incorporated the BoTNet module into the backbone network of the YOLO-v8 model. Enhancing the integrity of feature extraction by linking global and local features through its multi-head self-attention mechanism (MHSA). Additionally, a detection layer was added in the model’s neck structure with an output feature map scale of 160 × 160 to further integrate P. umbrosa feature details from the shallow neural network, thereby strengthening the recognition of small target P. umbrosa. The use of GSConv, as a replacement for some standard convolutions, not only reduced model computational complexity but also further improved its detection performance. Ablation test results reveal that the BSS-YOLOv8 network model achieved a precision of 91.1%, a recall rate of 86.7%, an mAP50 of 92.6%, an F1-Score of 88.85%, and an mAP50:95 of 61.3% on the P. umbrosa seedling dataset. Compared with the baseline network, it demonstrated respective improvements of 2.5%, 3.8%, 3.4%, 3.19%, and 4.4%. When compared to other object detection models (YOLO-v5, Faster R-CNN, etc.), the BSS-YOLOv8 model similarly achieved the best detection performance. The BSS-YOLOv8 proposed in this study enables rapid identification of P. umbrosa seedlings in grassland backgrounds, holding significant importance for early detection and control of weeds in grasslands.

A novel feature enhancement and semantic segmentation scheme for identifying low-contrast ocean oil spills

The oil spill accidents on the sea surface pose a severe threat to the marine environment and human health. This paper proposes a novel Semantic Segmentation Network (SSN) for processing oil spill images so that low-contrast oil spills on the sea surface can be accurately identified. After the detection accuracy and real-time performance of the current SSNs are compared, the basic network architecture of DeeplabV3+ based target detection is analyzed. The standard convolution is replaced by the Omni-dimensional Dynamic Convolution (ODConv) in the Ghost Module Depth-Wise separable Convolution (DWConv) to further enhance the feature extraction ability of the network. Furthermore, a new DeeplabV3+ based network with ODGhostNetV2 is constructed as the main feature extraction module, and an Adaptive Triplet Attention (ATA) module is deployed in the encoder and decoder at the same time. This not only improves the richness of semantic features but also increases the following receptive fields of the network model. ATA integrates the Adaptively Spatial Feature Fusion (ASFF) module to optimize the weight assignment problem in the feature map fusion process. The ablation experiments are conducted to verify the proposed network which show high accuracy and good real-time performance for the oil spill detection.

Dynamic context modeling based lightweight high-resolution network for dense prediction

Recent research shows that high-resolution networks can provide representative multi-scale features for vision tasks. However, high-resolution-based architectures are weak in capturing spatial long-range information, and they are computationally expensive. To alleviate these problems and encourage the engineering applications of dense prediction, based on vanilla high-resolution network, this paper presents a novel efficient architecture for dense prediction, namely Dynamic context modeling based lightweight High-Resolution Network (Dynamic-HRNet). In particular, the network consists of two key components: (i) dynamic split convolution, a more efficient and flexible convolution compared to the standard convolution, which could be conveniently embedded into other networks, and (ii) adaptive context modeling, which adaptively captures local and global contextual information to enrich the representation in parallel. Using the above two components, two lightweight blocks are designed that serve as the basic building units of Dynamic-HRNet, termed dynamic multi-scale context block and dynamic global context block, respectively. Extensive experiments demonstrate that the proposed Dynamic-HRNet, as a backbone, achieves significant superiority in popular benchmarks for human pose estimation (70.6% AP on COCO dataset and 87.6% PCKh on MPII dataset) and semantic segmentation (75.3% MIoU on Cityscapes dataset). Considering both efficiency and accuracy, it outperforms the most advanced lightweight architectures.

YOLO-SE: Improved YOLOv8 for Remote Sensing Object Detection and Recognition

Object detection remains a pivotal aspect of remote sensing image analysis, and recent strides in Earth observation technology coupled with convolutional neural networks (CNNs) have propelled the field forward. Despite advancements, challenges persist, especially in detecting objects across diverse scales and pinpointing small-sized targets. This paper introduces YOLO-SE, a novel YOLOv8-based network that innovatively addresses these challenges. First, the introduction of a lightweight convolution SEConv in lieu of standard convolutions reduces the network’s parameter count, thereby expediting the detection process. To tackle multi-scale object detection, the paper proposes the SEF module, an enhancement based on SEConv. Second, an ingenious Efficient Multi-Scale Attention (EMA) mechanism is integrated into the network, forming the SPPFE module. This addition augments the network’s feature extraction capabilities, adeptly handling challenges in multi-scale object detection. Furthermore, a dedicated prediction head for tiny object detection is incorporated, and the original detection head is replaced by a transformer prediction head. To address adverse gradients stemming from low-quality instances in the target detection training dataset, the paper introduces the Wise-IoU bounding box loss function. YOLO-SE showcases remarkable performance, achieving an average precision at IoU threshold 0.5 (AP50) of 86.5% on the optical remote sensing dataset SIMD. This represents a noteworthy 2.1% improvement over YOLOv8 and YOLO-SE outperforms the state-of-the-art model by 0.91%. In further validation, experiments on the NWPU VHR-10 dataset demonstrated YOLO-SE’s superiority with an accuracy of 94.9%, surpassing that of YOLOv8 by 2.6%. The proposed advancements position YOLO-SE as a compelling solution in the realm of deep learning-based remote sensing image object detection.

Deep Learning for Inversion of Tipper Data of a Certain Railway Tunnel in Tibet Area

The traditional inversion methods for tipper rely excessively on the selection of the initial model, whose global search capability is poor. Inspired by the significant approximation advantage of deep learning for nonlinear inverse problems with big data, we design a deep learning architecture called TipInv-net for the inversion of tipper. TipInv-net takes the improved U-net as the basic framework to obtain the tipper response characteristics of the abnormal body, and then, dense skip connection is applied among the nested standard convolution modules to alleviate the gradient disappearance problem and enhance feature propagation. It’s worth noting that we construct a feature pyramid of tipper response via average pooling to obtain multi-scale receptive fields, which, thus enhancing global and detail location of abnormal body. The theoretical model test indicates that the position and attitude of geological anomaly body can be distinguished via TipInv-net, even in the presence of a certain level of noise, the inversion accuracy will not be greatly affected. TipInv-net has strong generalization. Besides, after the training process of the network, we can obtain the inversion result immediately. In order to verify the effectiveness of the method in this paper, the inversion of the field tipper data of a section of a certain railway tunnel in Tibet area is carried out.

A Lightweight Multi-Scale Quadratic Separation Convolution Module for CNN Image-Classification Tasks

Currently, most convolutional networks use standard convolution for feature extraction to pursue accuracy. However, there is potential room for improvement in terms of the number of network parameters and model speed. Therefore, this paper proposes a lightweight multi-scale quadratic separable convolution module (Mqscm). First, the module uses a multi-branch topology to maintain the sparsity of the network architecture. Second, channel separation and spatial separation methods are used to separate the convolution kernels, reduce information redundancy within the network, and improve the utilization of hardware computing resources. In the end, the module uses a variety of convolution kernels to obtain information on different scales to ensure the performance of the network. The performance comparison on three image-classification datasets shows that, compared with standard convolution, the Mqscm module reduces computational effort by approximately 44.5% and the model training speed improves by a range of 14.93% to 35.41%, maintaining performance levels comparable to those of deep convolution. In addition, compared with ResNet-50, the pure convolution network MqscmNet reduces the parameters by about 59.5%, saves the training time by about 29.7%, and improves the accuracy by 0.59%. Experimental results show that the Mqscm module reduces the memory burden of the model, improves efficiency, and has good performance.

Group-Fusion One-Dimensional Convolutional Neural Network for Ballistic Target High-Resolution Range Profile Recognition with Layer-Wise Auxiliary Classifiers

Ballistic missile defense systems require accurate target recognition technology. Effective feature extraction is crucial for this purpose. The deep convolutional neural network (CNN) has proven to be an effective method for recognizing high-resolution range profiles (HRRPs) of ballistic targets. It excels in perceiving local features and extracting robust features. However, the standard CNN's fully connected manner results in high computational complexity, which is unsuitable for deployment in real-time missile defense systems with stringent performance requirements. To address the issue of computational complexity in HRRP recognition based on the standard one-dimensional CNN (1DCNN), we propose a lightweight network called group-fusion 1DCNN with layer-wise auxiliary classifiers (GFAC-1DCNN). GFAC-1DCNN employs group convolution (G-Conv) instead of standard convolution to effectively reduce model complexity. Simply using G-Conv, however, may decrease model recognition accuracy due to the lack of information flow between feature maps generated by each G-Conv. To overcome this limitation, we introduce a linear fusion layer to combine the output features of G-Convs, thereby improving recognition accuracy. Additionally, besides the main classifier at the deepest layer, we construct layer-wise auxiliary classifiers for different hierarchical features. The results from all classifiers are then fused for comprehensive target recognition. Extensive experiments demonstrate that GFAC-1DCNN with such simple and effective techniques achieves higher overall testing accuracy than state-of-the-art ballistic target HRRP recognition models, while significantly reducing model complexity. It also exhibits a higher recall rate for warhead recognition compared to other methods. Based on these compelling results, we believe this work is valuable in reducing workload and enhancing missile interception rates in missile defense systems.

A Lightweight Model for Real-Time Detection of Vehicle Black Smoke

This paper discusses the application of deep learning technology in recognizing vehicle black smoke in road traffic monitoring videos. The use of massive surveillance video data imposes higher demands on the real-time performance of vehicle black smoke detection models. The YOLOv5s model, known for its excellent single-stage object detection performance, has a complex network structure. Therefore, this study proposes a lightweight real-time detection model for vehicle black smoke, named MGSNet, based on the YOLOv5s framework. The research involved collecting road traffic monitoring video data and creating a custom dataset for vehicle black smoke detection by applying data augmentation techniques such as changing image brightness and contrast. The experiment explored three different lightweight networks, namely ShuffleNetv2, MobileNetv3 and GhostNetv1, to reconstruct the CSPDarknet53 backbone feature extraction network of YOLOv5s. Comparative experimental results indicate that reconstructing the backbone network with MobileNetv3 achieved a better balance between detection accuracy and speed. The introduction of the squeeze excitation attention mechanism and inverted residual structure from MobileNetv3 effectively reduced the complexity of black smoke feature fusion. Simultaneously, a novel convolution module, GSConv, was introduced to enhance the expression capability of black smoke features in the neck network. The combination of depthwise separable convolution and standard convolution in the module further reduced the model’s parameter count. After the improvement, the parameter count of the model is compressed to 1/6 of the YOLOv5s model. The lightweight vehicle black smoke real-time detection network, MGSNet, achieved a detection speed of 44.6 frames per second on the test set, an increase of 18.9 frames per second compared with the YOLOv5s model. The mAP@0.5 still exceeded 95%, meeting the application requirements for real-time and accurate detection of vehicle black smoke.

MS-YOLOv5: a lightweight algorithm for strawberry ripeness detection based on deep learning

The existing ripeness detection algorithm for strawberries suffers from low detection accuracy and high detection error rate. Considering these problems, we propose an improvement method based on YOLOv5, named MS-YOLOv5. The first step is to reconfigure the feature extraction network of MS-YOLOv5 by replacing the standard convolution with the depth hybrid deformable convolution (Ms-MDconv). In the second step, a double cooperative attention mechanism (Bc-attention) is constructed and implemented in the CSP2 module to improve the feature representation in complex environments. Finally, the Neck section of MS-YOLOv5 has been enhanced to use the fast-weighted fusion of cross-scale feature pyramid networks (FW-FPN) to replace the CSP2 module. It not only integrates multi-scale target features but also significantly reduces the number of parameters. The method was tested on the strawberry ripeness dataset, the mAP reached 0.956, the FPS reached 76, and the model size was 7.44M. The mAP and FPS are 8.4 and 1.3 percentage higher than the baseline network, respectively. The model size is reduced by 6.28M. This method is superior to mainstream algorithms in detection speed and accuracy. The system can accurately identify the ripeness of strawberries in complex environments, which could provide technical support for automated picking robots.

DCO-Net: Deformable convolution-based O-shape network for fully automated placenta segmentation

The placenta is an important organ that connects the fetus with the mother during pregnancy. Many pregnancies fail because of the placental abnormality. Nowadays, placental segmentation techniques based on magnetic resonance (MR) images are increasingly used to assist in the diagnosis of placental abnormalities. However, manual segmentation of the placenta is unreliable and time-consuming. In comparison to manually segmentation methods, the CNN-based methods perform better. However, due to the limitations of the fixed geometric structures in the traditional convolution, the features of the edge are not completely obtained. That is the challenge in placenta segmentation. By incorporating offsets into the regular grid sampling locations of standard convolution, deformable convolution allows for the utilization of spatial information to its full potential. Therefore, in this paper, we proposed a fully automatic placental segmentation method based on deformable convolution, termed as Deformable Convolution O-shape Network (DCO-Net). Specifically, we adopt the U-shaped architecture and employed a bi-directional O-shaped route by adding backward skip connections. The bi-directional O-shaped route recurrently reused the building blocks without introducing any extra parameters. To evaluate the performance of the proposed DCO-Net, a total of 871 MR images from 20 pregnant women were used for the experiments. The dice and average symmetric surface distance (ASSD) obtained by DCO-Net were 0.8488 and 3.0976, respectively. The experimental results demonstrate that the proposed DCO-Net is more efficient for automatic placental segmentation compared with the other competitive approaches.