Feature Reuse Research Articles

Automated pixel-level crack detection and quantification using deep convolutional neural networks for structural condition assessment

Crack detection is a crucial task in assessing the condition of concrete structures. Herein, a novel deep learning method based on convolutional neural networks, referred to as R-FPANet, is proposed for crack detection. The R-FPANet performs automatic segmentation and quantification of crack morphology at the pixel level. In this methodology, the modularization concept based on the following three modules is adopted: ResNet-50 is chosen as the backbone to extract features from images, the Feature Pyramid Network with Dense Block is integrated to promote the fusion of both shallow and deep features as well as enhance feature reuse, and self-attention mechanisms such as Channel Attention Module and Position Attention Module are introduced to strengthen the dependency between features. Based on the crack segmentation results, a suitably established framework is developed for quantitative analysis of the major geometric parameters, including crack area, crack length, crack mean width and crack max-width at the pixel level. To verify the effectiveness of the proposed method, a large-scale concrete crack image dataset was produced and carefully labeled at the pixel level and then utilized to train the model. Finally, our experiments reveal that the proposed approach achieves an Intersection over Union of 83.07%, further indicating that the segmentation performance of the proposed method is better than the state-of-the-art models and also confirming that the crack quantification results are close to reality. Overall, the proposed method performs well, contributing to crack detection and quantification with great potential for practical use.

Deep network with double reuses and convolutional shortcuts

AbstractThe authors design a novel convolutional network architecture, that is, deep network with double reuses and convolutional shortcuts, in which new compressed reuse units are presented. Compressed reuse units combine the reused features from the first 3 × 3 convolutional layer and the features from the last 3 × 3 convolutional layer to produce new feature maps in the current compressed reuse unit, simultaneously reuse the feature maps from all previous compressed reuse units to generate a shortcut by an 1 × 1 convolution, and then concatenate these new maps and this shortcut as the input to next compressed reuse unit. Deep network with double reuses and convolutional shortcuts uses the feature reuse concatenation from all compressed reuse units as the final features for classification. In deep network with double reuses and convolutional shortcuts, the inner‐ and outer‐unit feature reuses and the convolutional shortcut compressed from the previous outer‐unit feature reuses can alleviate the vanishing‐gradient problem by strengthening the forward feature propagation inside and outside the units, improve the effectiveness of features and reduce calculation cost. Experimental results on CIFAR‐10, CIFAR‐100, ImageNet ILSVRC 2012, Pascal VOC2007 and MS COCO benchmark databases demonstrate the effectiveness of authors’ architecture for object recognition and detection, as compared with the state‐of‐the‐art.

Automatic Plastic Greenhouse Extraction from Gaofen-2 Satellite Images with Fully Convolution Networks and Image Enhanced Index

Plastic greenhouses (PGs) play a vital role in modern agricultural development by providing a controlled environment for the cultivation of food crops. Their widespread adoption has the potential to revolutionize agriculture and impact the local environment. Accurate mapping and estimation of PG coverage are critical for strategic planning in agriculture. However, the challenge lies in the extraction of small and densely distributed PGs; this is often compounded by issues like irrelevant and redundant features and spectral confusion in high-resolution remote-sensing imagery, such as Gaofen-2 data. This paper proposes an innovative approach that combines the power of a full convolutional network (FC-DenseNet103) with an image enhancement index. The image enhancement index effectively accentuates the boundary features of PGs in Gaofen-2 satellite images, enhancing the unique spectral characteristics of PGs. FC-DenseNet103, known for its robust feature propagation and extensive feature reuse, complements this by addressing challenges related to feature fusion and misclassification at the boundaries of PGs and adjacent features. The results demonstrate the effectiveness of this approach. By incorporating the image enhancement index into the DenseNet103 model, the proposed method successfully eliminates issues related to the fusion and misclassification of PG boundaries and adjacent features. The proposed method, known as DenseNet103 (Index), excels in extracting the integrity of PGs, especially in cases involving small and densely packed plastic sheds. Moreover, it holds the potential for large-scale digital mapping of PG coverage. In conclusion, the proposed method providing a practical and versatile tool for a wide range of applications related to the monitoring and evaluation of PGs, which can help to improve the precision of agricultural management and quantitative environmental assessment.

Uncertainty-driven trustworthy defect detection for high-resolution powder bed images in selective laser melting

Selective laser melting (SLM) is known as one of the most promising metal additive manufacturing technologies, and how to ensure its consistent quality is still a main challenge, which urgently needs to be addressed. The metal powder spreading quality, as the first stage of SLM, has a direct impact on the subsequent forming and it is necessary to be monitored. However, existing deep learning-based methods for monitoring powder spreading quality often suffer from the problem of unreliability. The high-resolution property of powder bed images is often ignored, and the predicted results are not well evaluated. To address the above issues and achieve trustworthy defect detection, this paper proposes an uncertainty-driven trustworthy defect detection method for high-resolution powder bed images in SLM. A super resolution module based on ISDNet is first proposed to refine the upsampling process. Checkpoint ensemble is adopted for better achieving model uncertainty estimation only requiring a single training process, feature reuse helps reduce the computational load, and temperature scaling is used to further calibrate each ensemble member to get a more precise uncertainty. Besides, we design an uncertainty-driven model improvement method to further improve defect detection performance for regions with high uncertainty. Experiments demonstrate the effectiveness of our proposed method, and our work achieves a trustworthy defect detection for high-resolution powder bed images in SLM.

Basis scaling and double pruning for efficient inference in network-based transfer learning

Network-based transfer learning allows the reuse of deep learning features with limited data, but the resulting models can be unnecessarily large. Although network pruning can improve inference efficiency, existing algorithms usually require fine-tuning that may not be suitable for small datasets. In this paper, using the singular value decomposition, we decompose a convolutional layer into two layers: a convolutional layer with the orthonormal basis vectors as the filters, and a “BasisScalingConv” layer which is responsible for rescaling the features and transforming them back to the original space. As the filters in each decomposed layer are linearly independent, when using the proposed basis scaling factors with the Taylor approximation of importance, pruning can be more effective and fine-tuning individual weights is unnecessary. Furthermore, as the numbers of input and output channels of the original convolutional layer remain unchanged after basis pruning, it is applicable to virtually all architectures and can be combined with existing pruning algorithms for double pruning to further increase the pruning capability. When transferring knowledge from ImageNet pre-trained models to different target domains, with less than 1% reduction in classification accuracies, we can achieve pruning ratios up to 74.6% for CIFAR-10 and 98.9% for MNIST in model parameters.

PCB Defect Image Recognition Based On The Multi-Model Fusion Algorithm

Abstract Printed Circuit Board (PCB) is one of the most important components of electronic products. But the traditional defect detection methods are gradually difficult to meet the requirements of PCB defect detection. The research on PCB defect recognition method based on convolutional neural network is the current trend. The PCB defect image recognition based on DenseNet169 network model is studied in this paper. In order to reduce the omission of PCB defects in actual detection, it is necessary to further improve the sensitivity of the model. Therefore, a classification model based on the multi-model fusion of the DenseNet169 model and the ResNet50 model is proposed. At the same time, the network structure after multi-model fusion is improved. The improved multi-model fusion model Mix-Fusion enables the network to not only retain the recognition accuracy of the ResNet50 model for NG defects and small defect images, but also improve the overall recognition accuracy through the feature reuse and bypass settings of the DenseNet169 model. The experimental results show that when the threshold is 0.5, the sensitivity of the improved multi-model fusion network can reach 99.2%, and the specificity is 99.5%. The sensitivity of Mix-Fusion is 1.2% higher than that of DenseNet169. High sensitivity means fewer missed NG images, and high specificity means less workload for employees. The improved model improves sensitivity and maintains high specificity.

Hybrid augmented network with balance domain window for few-shot fault diagnosis under sharp speed variation

In engineering practice, the operation of mechanical equipment under conditions of sharp speed variation can result in domain shift in the distribution of samples. Moreover, the presence of data deficiencies in engineering poses significant challenges for intelligent diagnostic techniques. To address these issues, this paper proposes a network called Hybrid Augmented network with Balance Domain Window (BW-HAN). The BW-HAN network utilizes convolution operations to segment samples into infinitesimal patches while extracting their underlying features, and then embeds these patches into attention computational mechanism for invariant domain feature extraction. To handle domain shift, a novel partitioning method is designed based on data restructuring and data flow. This method aims to suppress domain shift within windows while establishing global connectivity between windows. Furthermore, a semi-dense connection method with multi-level residual fusion has been developed based on the principles of feature reuse and parameter sharing to address the issue of overfitting caused by limited samples, thereby enhancing model stability. Comparative studies are conducted with other networks on three variable-speed cases to demonstrate the superiority of the BW-HAN network. The experimental results show that the proposed BW-HAN network performs well in all three cases and effectively addresses the problem of insufficient data under sharp speed variation in mechanical fault diagnosis.

Screen-shot and Demoiréd image identification based on DenseNet and DeepViT

How to identify the screen-shot image is an important branch of image source forensics. Although the physical feature of moiré patterns may be left after LCD recapture operation, this feature may also be well concealed with the development of deep learning-based techniques for demoiréing operation. In this paper, to solve the forensic problem of screen-shot and demoiréd image operation, we select several typical demoiréing algorithms to generate datasets for training and detecting demoiréd images. Meanwhile, we also perform the task of identifying specific demoiréing algorithms. DenseNet can extract fine demoiréd image features because of its enhanced feature reuse and improved model performance. The re-attention mechanism is also added to our designed network to extract the image’s global features and avoid the self-attention collapse problem in ViT. Therefore, we design a network based on DenseNet cascaded with DeepViT structure to detect the demoiréd image and identify its corresponding demoiréing algorithm. In a large number of comparison experiments of various typical image recapture detection methods, our method achieves the best detection results and network performance.

Soybean Leaf Diseases Recognition Based on Generative Adversarial Network and Transfer Learning

Soybean leaf disease labeling data are not easy to obtain, and soybean leaf disease model training often needs a lot of data. Due to the limitations of fixed rules such as rotation and clipping, traditional data enhancement cannot generate images with diversity and variability. In view of the above problems, this study proposed a data enhancement method based on generative adversarial network to expand the original soybean leaf disease dataset. This method was based on cyclic confrontation network, and its discriminator uses dense connection strategy to realize feature reuse, so as to reduce the amount of calculation. In the training process, improved transfer learning is used to automatically fine tune the pre-training model. The accuracy of the optimized method in 9 kinds of soybean leaf disease image recognition is 95.84%, which is 0.98% higher than the traditional fine-tuning method. The experimental results show that this method based on generating confrontation network has significant ability in generating soybean leaf disease image, and can expand the existing dataset. In addition, this method also provides an effective data enhancement solution for the expansion of other crop disease image datasets.

An efficient joint framework assisted by embedded feature smoother and sparse skip connection for hyperspectral image classification

Deep neural network (DNN) methods play an essential role in hyperspectral classification. However, the massive parameters and vast computing overhead of DNN needs to be reduced when facing the deployment with limited storage and computing resources for real-time response applications, especially considering the high dimensionality of hyperspectral image. So, applying dimension reduction (DR) methods is a crucial pre-processing method in various studies. Still, most of them ignore the feature restoration after the data transformation by DR. In neural networks, many works still involve sophisticated skip connections and dense feature reuse, which can lead to feature redundancy and increase computational complexity, especially when DR methods have been applied first. Motivated by these issues, an efficient joint framework assisted by embedded feature smoother (FS) and sparse skip connection (SSC) is proposed in this article. Instead of directly feeding DR data into the subsequent network, we embedded a computing-cheap FS based on isotropic total variation to restore and enhance the spatial features. Furthermore, we proposed a SSC 3D convolution neural network to complete spatial–spectral feature representation and classification. The SSC is embodied in the design of log2n-skip connection to concatenate feature maps instead of dense connection, pruning the number of channels and reducing the model parameters. Experimental results show that the embedded FS significantly improves classification accuracy and is superior to other processing methods. Our framework offers much superior to other state-of-the-art deep learning-based methods considering classification performance and lightweight aspects, especially when using few training samples. Moreover, considering detailed processing steps, our framework has a competitively cheaper time consumption.

Micro-Expression Spotting Based on VoVNet, Driven by Multi-Scale Features

Micro-expressions are a type of real emotional expression, which are unconscious and difficult to hide. Identifying these expressions has great potential applications in areas such as civil aviation security, criminal interrogation, and clinical medicine. However, because of their characteristics such as short duration, low intensity, and sparse action units, this makes micro-expression spotting difficult. To address this problem and inspired by object detection methods, we propose a VoVNet-based micro-expression spotting model, driven by multi-scale features. Firstly, VoVNet is used to achieve the extraction and reuse of different scale perceptual field features to improve the feature extraction capability. Secondly, multi-scale features are extracted and fused using the Feature Pyramid Network module, incorporating optical flow features, and by realizing the interactive fusion of fine-grained feature information and semantic feature information. Finally, the model is trained and optimized on CAS(ME)2 and SAMM Long Video. The experimental results show that the F1 score of the proposed model is improved by 0.1963 and 0.2441 on the two datasets compared with the baseline method, which outperforms the most popular spotting methods.

Feature recognition in multiple CNNs using sEMG images from a prototype comfort test

ObjectiveDeep learning-based CNN networks have recently been investigated to solve the problem of body posture recognition based on surface electromyographic signals (sEMG). Influenced by these studies, to develop a combined approach of sEMG and CNNs in the study of human-product interactions and the impact of body comfort, and to compare the advantages and disadvantages of various CNNs networks. MethodsIn this study, sEMG measurements were carried out by building a prototype usability experiment, and the data were divided into four categories, with two types of datasets: training and testing. Four CNNs, LeNet-5, VGGNet-11, InceptionNet V4, and DenseNet, were used for the recognition of sEMG images. ResultsDenseNet is another type of convolutional neural network with deep layers, which has a unique advantage over other algorithms. unique advantages over other algorithms. DenseNet has fewer layers and better accuracy than InceptionNet V4, but not only does it bypass enhanced feature reuse, but its network is easier to train and has some regularization effects, while also mitigating the problems of gradient disappearance and model degradation. ConclusionThese findings could lead to a more appropriate CNN model and a useful tool for developing comfort judgments of surface EMG signals, furthering the development of products that come into contact with the human body without the need for routine retraining.

Radiologically based automated segmentation of cardiac MRI using an improved U-Net neural algorithm

ObjectiveTo overcome the complexity and variability of the heart, along with factors like fuzzy boundaries and low contrast and produce automated segmentation based on machine learning. MethodsSegmentation of cardiac MRI plays a vital role in various clinical applications. In this study, we propose an improved U-Net-CSP (Cross Stage Partial) method for accurate and robust segmentation of cardiac MRI images.The U-Net-CSP architecture combines the U-Net framework with the CSP module to enhance the segmentation performance. The CSP module enables improved feature reuse and mitigates overfitting issues associated with deep networks. By incorporating the CSP module in both the encoding and decoding stages, the U-Net-CSP effectively captures complex features and preserves important details during the upsampling and downsampling processes.To optimize the segmentation results, we utilize a combination of weighted cross-entropy loss, dice loss, and L2 regularization as the loss function. This loss function facilitates accurate delineation of the regions of interest and enhances the segmentation performance of the U-Net-CSP model. ResultsThe proposed U-Net-CSP method achieves superior performance in cardiac MRI segmentation, surpassing conventional U-Net models and state-of-the-art methods. It demonstrates potential for clinical applications in cardiac therapy, disease research, and model reconstruction. Our method automates segmentation of the left ventricle, right ventricle, and aorta from cardiac MRI images. Comparative analyses with leading methods show consistent favorable outcomes under diverse conditions. The U-Net-CSP approach accurately delineates cardiac structures from MRI scans, contributing to medical image analysis and advancing cardiac healthcare. ConclusionIn conclusion, the proposed U-Net-CSP method provides an enhanced approach for cardiac MRI segmentation, overcoming the challenges posed by complex cardiac structures and image characteristics. This study contributes to the advancement of medical image analysis techniques, particularly in the field of cardiac MRI segmentation.

DP-FishNet: Dual-path Pyramid Vision Transformer-based underwater fish detection network

The detection of underwater fish targets is critical for ecological monitoring and marine biodiversity research. However, underwater fish detection is typically constrained by problems including low image quality and variable underwater surroundings. On behalf of further improving the underwater fish detection accuracy in complex underwater environments, this paper proposes a dual-path (DP) Pyramid Vision Transformer (PVT) feature extraction network named DP-FishNet. The backbone network DP-PVT composed from the PVT Network is made up of two feature extraction paths. The first represents the Vision Transformer path, which extracts global features to enhance the distinction between the foreground and background of underwater images. The second is the convolutional neural network path, which enhances the accuracy of detecting small targets by extracting local features. Additionally, to more effectively utilize the feature information extracted by the network, this paper provides a promising solution to employ the content-aware reassembly of features (Carafe) in the feature pyramid network (FPN). The seesaw loss is utilized as a classification loss to address the problem of unbalanced samples caused by the gap in the number of fish populations. According to the experimental findings, the AP and AP50 of the DP-FishNet are 76.0% and 95.2%, respectively. In comparison to currently available advanced two-stage detection algorithms, the quantity of computation and parameters is reduced by approximately 40%. DP-FishNet strengthens the ability to extract global and local features from underwater images and enhances feature reuse. DP-FishNet can be utilized to detect fish targets in actual and complicated underwater habitats.

Topology structure optimization of evolutionary hierarchical fuzzy systems

In this paper, a new method of hierarchical fuzzy system modeling for high-dimensional regression problems is proposed, which is called multi-objective evolutionary hierarchical fuzzy regression system (MOEHFRS). Different from the existing hierarchical fuzzy systems with fixed topology, the proposed MOEHFRS improves the accuracy of the model, reduces the total number of rules, and eliminates unnecessary features by flexibly constructing topology. In the process of topology evolution, MOEHFRS can exchange and combine different sub-fuzzy systems to achieve the selection and reuse of important features, as well as the elimination of unnecessary features, which improves the diversity of topology and enables the model to be well applied to high-dimensional regression problems. Different combinations of sub-fuzzy systems will result in different performance and number of rules. A new multi-objective evolutionary optimization algorithm is proposed to simultaneously optimize the number of rules and the accuracy of the model, which achieves the balance between complexity and accuracy of MOEHFRS. The proposed method is validated on 13 real world regression datasets and compared with other 5 methods. The results show that MOEHFRS is effective and advanced in terms of accuracy, number of rules, retention of important features and universality in different regression problems.

ResFuseYOLOv4_Tiny: Enhancing detection accuracy for lightweight networks in infrared small object detection tasks

When deploying infrared target detection systems at the edge, one often encounters constraints on computational and storage resources. In such scenarios, lightweight networks prove to be a viable detection solution. Compared to large-scale deep networks, lightweight networks, after discarding redundant structures, often do not achieve satisfactory detection accuracy, especially in the context of small object detection tasks. This paper proposes ResFuseYOLOv4_Tiny, an improved lightweight network specifically designed for the effective detection of infrared small objects, particularly in resource-constrained scenarios. We extend the YOLOv4_Tiny neck network to enable efficient feature extraction. By integrating Res_head with residual structures and fusion-based feature reuse, our framework substantially enhances the capability to detect small objects by expanding the receptive field and effectively merging features from various receptive fields. Additionally, we introduce a drop and adaptive parameter-based attention mechanism called DCA, leveraging the inclusion of drop layers to effectively reduce information redundancy. Moreover, we replace the conventional CIoU loss function with our customized MSIoU, incorporating parameter m to adaptively adjust the loss weights in response to the influence of small target sizes. Experimental results conducted on synthetic and real infrared small object datasets respectively reveal a significant 14.47% and 11.28% improvement in detection accuracy. The proposed ResFuseYOLOv4_Tiny framework presents a viable solution for lightweight networks when both detection speed and accuracy are concurrently considered.

Automated stratigraphic correlation of well logs using Attention Based Dense Network

The stratigraphic correlation of well logs plays an essential role in characterizing subsurface reservoirs. However, it suffers from a small amount of training data and expensive computing time. In this work, we propose the Attention Based Dense Network (ASDNet) for the stratigraphic correlation of well logs. To implement the suggested model, we first employ the attention mechanism to the input well logs, which can effectively generate the weighted well logs to serve for further feature extraction. Subsequently, the DenseNet is utilized to achieve good feature reuse and avoid gradient vanishing. After model training, we employ the ASDNet to the testing data set and evaluate its performance based on the well log data set from Northwest China. Finally, the numerical results demonstrate that the suggested ASDNet provides higher prediction accuracy for automated stratigraphic correlation of well logs than state-of-the-art contrastive UNet and SegNet.

MDAU-Net: A Liver and Liver Tumor Segmentation Method Combining an Attention Mechanism and Multi-Scale Features

In recent years, U-Net and its extended variants have made remarkable progress in the realm of liver and liver tumor segmentation. However, the limitations of single-path convolutional operations have hindered the full exploitation of valuable features and restricted their mobility within networks. Moreover, the semantic gap between shallow and deep features proves that a simplistic shortcut is not enough. To address these issues and realize automatic liver and tumor area segmentation in CT images, we introduced the multi-scale feature fusion with dense connections and an attention mechanism segmentation method (MDAU-Net). This network leverages the multi-head attention (MHA) mechanism and multi-scale feature fusion. First, we introduced a double-flow linear pooling enhancement unit to optimize the fusion of deep and shallow features while mitigating the semantic gap between them. Subsequently, we proposed a cascaded adaptive feature extraction unit, combining attention mechanisms with a series of dense connections to capture valuable information and encourage feature reuse. Additionally, we designed a cross-level information interaction mechanism utilizing bidirectional residual connections to address the issue of forgetting a priori knowledge during training. Finally, we assessed MDAU-Net’s performance on the LiTS and SLiver07 datasets. The experimental results demonstrated that MDAU-Net is well-suited for liver and tumor segmentation tasks, outperforming existing widely used methods in terms of robustness and accuracy.

Unsupervised transfer learning for intelligent health status identification of bearing in adaptive input length selection

Input length (IL) is an important element in transfer learning (TL) network for intelligent health status identification of bearing (IHSIB). However, fixed IL are used in most studies. In this paper, a TL network via adaptive IL selection module for IHSIB (AILTLN) is proposed, which includes adaptive IL module, feature extractor module, health status identification module, and domain discriminator module. Firstly, an adaptive IL selection module based on envelope spectrum analysis is proposed. The module varies with bearing structure, motor speed, and sampling frequency. Secondly, group convolution, transposed convolution, and instant normalization are constructed in feature extractor. Thirdly, softmax cross-entropy loss function and maximum mean discrepancy are used for health status identification and domain alignment. The TL results of open bearing dataset and high-speed train bearing experiment show that AILTLN is better than the other existing methods in the TL of IHSIB. The ablation study shows the reuse of low-dimensional features and the adaptive IL help to improve to the accuracy of the proposed method.

Weighted target feature regression neural networks based soft sensing for industrial process

AbstractThere are often nonlinear and time‐varying characteristics in industrial processes. These characteristics cause difficulty in measuring product quality online. To address these issues, a weighted target feature regression neural network (WTFAER) was proposed for soft sensor modelling in this paper. The Pearson correlation coefficient was calculated to assign corresponding weights to process variables and design a weighted objective function. A target feature regression network (TFAER) was constructed using target correlation autoencoder with fully connected layer. After that, the weighted reconstructed information was applied to the TFAER model to extract deep quality‐related features and realize feature reuse. A deep network was formed by layered stacking to fully exploit the deep features for quality prediction. To make the proposed method domain adaptive, a maximum mean squared deviation (MMD) based regularization term was introduced in the loss function. Through the simulation experiments of debutanizer column and industrial polyethylene process, and compared with stacked autoencoder (SAE), variable‐wise weighted stacked autoencoder (VW‐SAE) and stacked target‐related autoencoder (STAE) methods, the effectiveness and generalization performance of the proposed modelling method were verified.