Features Of Different Scales Research Articles

Sea ice image classification based on ResFPG network and heterogeneous data fusion

ABSTRACT Sea ice detection has played an important role in climate protection and strategic deployment. Due to the autonomous learning characteristics of deep learning, it has been gradually applied to the classification of remote sensing sea ice images. At present, deep learning models are mostly used in sea ice classification of single source remote sensing data. Due to the limitations of single source data and the information loss of deep learning model in the process of feature extraction layer by layer, it is inevitable to encounter a bottleneck in sea ice detection requiring fine classification. To solve the above problems, this paper proposes a sea ice image classification method based on ResNet16-feature pyramid networks-spatial pyramid pooling-gated fusion network (ResFPG) and heterogeneous data fusion. In the feature extraction part, the method uses the improved ResNet16 to extract the multi-level feature information of sea ice in synthetic aperture radar data and optical data, reducing information loss in the feature extraction process, then mines and fuses the low-level spatial information and high-level semantic information through the improved feature pyramid networks (FPN), and then collects and fuses the output features of different scales through the spatial pyramid pooling (SPP) network. In the feature fusion part, a gated feature-level fusion strategy is designed to further improve the overall classification accuracy by adaptively adjusting the feature contribution of two heterogeneous sources of data through the gated fusion network (GFN). In order to verify the effectiveness of this method, we use two sets of heterogeneous sea ice remote sensing data located in Hudson Bay area for experiments. The experimental results show that compared with other image classification methods, the proposed method fully excavates and integrates the multi-scale and multi-level features in heterogeneous data, effectively distinguishes the feature contribution, and achieves better classification results (97.14% and 95.85%).

Siamese multiscale residual feature fusion network for aero-engine bearing fault diagnosis under small-sample condition

Implementing condition monitoring and fault diagnosis of aero-engine bearings is crucial to ensure that aircraft operate safely and reliably. In engineering practice, the fault data for aero-engine bearings are extremely limited. However, the traditional fault diagnosis methods have two shortcomings under extremely small sample conditions: (1) they have limited diagnostic performance and generalization ability, and (2) they do not mine fault information sufficiently or efficiently. This article proposes a Siamese multiscale residual feature fusion network (SMSRFFN) for aero-engine bearing fault diagnosis under small-sample conditions to overcome the weaknesses above. In the proposed SMSRFFN, the training samples are first paired according to the matching rules to realize the expansion of the sample size. Second, a multiscale residual feature extraction network (MSRFEN) is constructed to excavate the fault features of different scales and speed up the convergence speed of the network. Then, a multiscale attention mechanism feature fusion module (MSAMFFM) is designed to achieve efficient fusion of fault features at different scales. Finally, the distance of the input sample is measured based on the fused deep feature representation to identify the fault state of the aero-engine bearing. The proposed SMSRFFN is evaluated using three bearing fault data and also compared with some state-of-the-art small-sample diagnostic methods. The experimental results demonstrate the effectiveness and superiority of the proposed SMSRFFN in mining fault information and improving diagnosis accuracy under extremely small sample conditions.

Research on Fault Diagnosis Method Based on Improved CNN

Traditional fault diagnosis methods require complex signal processing and expert experience, and the accuracy of fault identification is low. To solve these problems, a fault diagnosis method based on an improved convolutional neural network (CNN) is proposed. Based on the traditional CNN model, a variety of convolution stride modes were added to extract features of different scales of signals and expand the feature dimension. Firstly, the vibration signals were collected and grouped. Then, the data were divided into a training set and a test set and input into improved CNN for feature extraction and model training to realize fault identification. The proposed model achieved a classification accuracy of 99.3% when testing the vibration data of the armored vehicle. Finally, the proposed model was used to classify different fault types of planetary gearboxes. The gradient-weighted class activation mapping (Grad-CAM) method was used to visualize the classification weight of samples. The results showed that the classification accuracy reaches 98% under various working conditions of the planetary gearbox.

Transformer-Based Model with Dynamic Attention Pyramid Head for Semantic Segmentation of VHR Remote Sensing Imagery.

Convolutional neural networks have long dominated semantic segmentation of very-high-resolution (VHR) remote sensing (RS) images. However, restricted by the fixed receptive field of convolution operation, convolution-based models cannot directly obtain contextual information. Meanwhile, Swin Transformer possesses great potential in modeling long-range dependencies. Nevertheless, Swin Transformer breaks images into patches that are single-dimension sequences without considering the position loss problem inside patches. Therefore, Inspired by Swin Transformer and Unet, we propose SUD-Net (Swin transformer-based Unet-like with Dynamic attention pyramid head Network), a new U-shaped architecture composed of Swin Transformer blocks and convolution layers simultaneously through a dual encoder and an upsampling decoder with a Dynamic Attention Pyramid Head (DAPH) attached to the backbone. First, we propose a dual encoder structure combining Swin Transformer blocks and reslayers in reverse order to complement global semantics with detailed representations. Second, aiming at the spatial loss problem inside each patch, we design a Multi-Path Fusion Model (MPFM) with specially devised Patch Attention (PA) to encode position information of patches and adaptively fuse features of different scales through attention mechanisms. Third, a Dynamic Attention Pyramid Head is constructed with deformable convolution to dynamically aggregate effective and important semantic information. SUD-Net achieves exceptional results on ISPRS Potsdam and Vaihingen datasets with 92.51%mF1, 86.4%mIoU, 92.98%OA, 89.49%mF1, 81.26%mIoU, and 90.95%OA, respectively.

Stacked attention hourglass network based robust facial landmark detection

Deep learning based facial landmark detection (FLD) has made rapid progress. However, the accuracy and robustness of FLD algorithms are degraded heavily when the face is subject to diverse expressions, posture deflection, partial occlusion and other uncertain circumstances. To learn more discriminative representations and reduce the negative effect caused by outliers, a stacked attention hourglass network (SAHN) is proposed for FLD, where new attention mechanism is introduced. Basically, in the design of SAHN, a spatial attention residual (SAR) unit is constructed such that relevant areas of facial landmarks are specially emphasized and essential features of different scales can be well extracted, and a channel attention branch (CAB) is introduced to better guide the next-level hourglass network for feature extraction. Due to the introduction of SAR and CAB, only two hourglass networks are stacked as the proposed SAHN with fewer parameters, which is different from traditional SHNs stacked by four hourglass networks. Furthermore, a variable robustness (VR) loss function is introduced for the training of SAHN. The robustness of the proposed model for FLD is guaranteed with the help of the VR loss by adaptively adjusting a continuous parameter. Extensive experimental results on three public datasets including 300W, WFLW and COFW confirm that our method is superior to some previous ones.

Detection Method of Marine Biological Objects Based on Image Enhancement and Improved YOLOv5S

Marine biological object detection is of great significance for the exploration and protection of underwater resources. There have been some achievements in visual inspection for specific objects based on machine learning. However, owing to the complex imaging environment, some problems, such as low accuracy and poor real-time performance, have appeared in these object detection methods. To solve these problems, this paper proposes a detection method of marine biological objects based on image enhancement and YOLOv5S. Contrast-limited adaptive histogram equalization is taken to solve the problems of underwater image distortion and blur, and we put forward an improved YOLOv5S to improve accuracy and real-time performance of object detection. Compared with YOLOv5S, coordinate attention and adaptive spatial feature fusion are added in the improved YOLOv5S, which can accurately locate the target of interest and fully fuse the features of different scales. In addition, soft non-maximum suppression is adopted to replace non-maximum suppression for the improvement of the detection ability for overlapping objects. The experimental results show that the contrast-limited adaptive histogram equalization algorithm can effectively improve the underwater image quality and the detection accuracy. Compared with the original model (YOLOv5S), the proposed algorithm has a higher detection accuracy. The detection accuracy AP50 reaches 94.9% and the detection speed is 82 frames per second; therefore, the real-time performance can be said to reach a high level.

PODD: A Dual-Task Detection for Greenhouse Extraction Based on Deep Learning

The rapid boom of the global population is causing more severe food supply problems. To deal with these problems, the agricultural greenhouse is an effective way to increase agricultural production within a limited space. To better guide agricultural activities and respond to future food crises, it is important to obtain both the agricultural greenhouse area and quantity distribution. In this study, a novel dual-task algorithm called Pixel-based and Object-based Dual-task Detection (PODD) that combines object detection and semantic segmentation is proposed to estimate the quantity and extract the area of agricultural greenhouses based on RGB remote sensing images. This algorithm obtains the quantity of agricultural greenhouses based on the improved You Only Look Once X (YOLOX) network structure, which is embedded with Convolutional Block Attention Module (CBAM) and Adaptive Spatial Feature Fusion (ASFF). The introduction of CBAM can make up for the lack of expression ability of its feature extraction layer to retain more important feature information. Adding the ASFF module can make full use of the features in different scales to increase the precision. This algorithm obtains the area of agricultural greenhouses based on the DeeplabV3+ neural network using ResNet-101 as a feature extraction network, which not only effectively reduces hole and plaque issues but also extracts edge details. Experimental results show that the mAP and F1-score of the improved YOLOX network reach 97.65% and 97.50%, 1.50% and 2.59% higher than the original YOLOX solution. At the same time, the accuracy and mIoU of the DeeplabV3+ network reach 99.2% and 95.8%, 0.5% and 2.5% higher than the UNet solution. All of the metrics in the dual-task algorithm reach 95% and even higher. Proving that the PODD algorithm could be useful for agricultural greenhouse automatic extraction (both quantity and area) in large areas to guide agricultural policymaking.

Shift Pooling PSPNet: Rethinking PSPNet for Building Extraction in Remote Sensing Images from Entire Local Feature Pooling

Building extraction by deep learning from remote sensing images is currently a research hotspot. PSPNet is one of the classic semantic segmentation models and is currently adopted by many applications. Moreover, PSPNet can use not only CNN-based networks but also transformer-based networks as backbones; therefore, PSPNet also has high value in the transformer era. The core of PSPNet is the pyramid pooling module, which gives PSPNet the ability to capture the local features of different scales. However, the pyramid pooling module also has obvious shortcomings. The grid is fixed, and the pixels close to the edge of the grid cannot obtain the entire local features. To address this issue, an improved PSPNet network architecture named shift pooling PSPNet is proposed, which uses a module called shift pyramid pooling to replace the original pyramid pooling module, so that the pixels at the edge of the grid can also obtain the entire local features. Shift pooling is not only useful for PSPNet but also in any network that uses a fixed grid for downsampling to increase the receptive field and save computing, such as ResNet. A dense connection was adopted in decoding, and upsampling was gradually carried out. With two open datasets, the improved PSPNet, PSPNet, and some classic image segmentation models were used for comparative experiments. The results show that our method is the best according to the evaluation metrics, and the predicted image is closer to the label.

Improving Hybrid CTC/Attention Architecture for Agglutinative Language Speech Recognition.

Unlike the traditional model, the end-to-end (E2E) ASR model does not require speech information such as a pronunciation dictionary, and its system is built through a single neural network and obtains performance comparable to that of traditional methods. However, the model requires massive amounts of training data. Recently, hybrid CTC/attention ASR systems have become more popular and have achieved good performance even under low-resource conditions, but they are rarely used in Central Asian languages such as Turkish and Uzbek. We extend the dataset by adding noise to the original audio and using speed perturbation. To develop the performance of an E2E agglutinative language speech recognition system, we propose a new feature extractor, MSPC, which uses different sizes of convolution kernels to extract and fuse features of different scales. The experimental results show that this structure is superior to VGGnet. In addition to this, the attention module is improved. By using the CTC objective function in training and the BERT model to initialize the language model in the decoding stage, the proposed method accelerates the convergence of the model and improves the accuracy of speech recognition. Compared with the baseline model, the character error rate (CER) and word error rate (WER) on the LibriSpeech test-other dataset increases by 2.42% and 2.96%, respectively. We apply the model structure to the Common Voice—Turkish (35 h) and Uzbek (78 h) datasets, and the WER is reduced by 7.07% and 7.08%, respectively. The results show that our method is close to the advanced E2E systems.

Extraction of Rural Residential Land from Very-High Resolution UAV Images Using a Novel Semantic Segmentation Framework

Accurate recognition and extraction of rural residential land (RRL) is significant for scientific planning, utilization, and management of rural land. Very-High Resolution (VHR) Unmanned Aerial Vehicle (UAV) images and deep learning techniques can provide data and methodological support for the target. However, RRL, as a complex land use assemblage, exhibits features of different scales under VHR images, as well as the presence of complex impervious layers and backgrounds such as natural surfaces and tree shadows in rural areas. It still needs further research to determine how to deal with multi-scale features and accurate edge features in such scenarios. In response to the above problems, a novel framework named cascaded dense dilated network (CDD-Net), which combines DenseNet, ASPP, and PointRend, is proposed for RRL extraction from VHR images. The advantages of the proposed framework are as follows: Firstly, DenseNet is used as a feature extraction network, allowing feature reuse and better network design with fewer parameters. Secondly, the ASPP module can better handle multi-scale features. Thirdly, PointRend is added to the model to improve the segmentation accuracy of the edges. The research takes a plain village in China as the research area. Experimental results show that the Precision, Recall, F1 score, and Dice coefficients of our approach are 91.41%, 93.86%, 92.62%, and 0.8359, respectively, higher than other advanced models used for comparison. It is feasible in the task of high-precision extraction of RRL using VHR UAV images. This research could provide technical support for rural land planning, analysis, and formulation of land management policies.

A Fine-Grained Image Classification and Detection Method Based on Convolutional Neural Network Fused with Attention Mechanism.

Due to the existence of attention system, people pay attention to the distinguishable area of the image, rather than directly receiving and processing the information of the whole image. This natural advantage makes attention mechanism widely used in fine-grained image classification. The research goal of fine-grained image classification task often is to differentiate subclass objects belonging to the same basic category. The difficulty of classification is that there are only slight local differences between different categories, but there may be large feature differences within the same category. At the same time, complex background features also bring interference factors to image recognition. In order to further extract discriminant regional features, this paper proposes a fine-grained image classification method WSFF-BCNN based on weak supervision feature fusion from two aspects: the improvement of the loss function in the training process of convolution neural network and the refinement of fine-grained image feature extraction. It uses the mixed attention of channel domain and spatial domain to obtain the detailed description information in the feature to highlight the response of the corresponding channel and spatial location in the feature map and pay attention to the attention characteristics of different dimensions. The original images of different sizes are input into the improved bilinear model to obtain multi-scale features. The large-scale features can represent the spatial location information of key areas, and the small-scale features represent the low-level features of the image. The backbone network of bilinear network uses ResNet50 to extract features and sample and zoom and uses bilinear pooling to fuse features of different scales to obtain a rich image feature representation.

Multi-scale long-range interactive and regional attention network for stroke lesion segmentation

High-performance segmentation can help physicians complete clinical diagnosis in a timely manner, thereby determining critical treatment periods and improving the efficiency of stroke diagnosis. However, such a lesion segmentation suffers from severe deviation and time-consuming due to small discrepancies between the lesions and healthy tissues. In order to segment stroke lesions with high performance, we intend to capture several scales of local–global semantic information in this study. Here, we propose the multi-scale long-range interactive and regional attention network (MLiRA-Net), which not only employs convolutional layers to build local features by patch partition block, but also adopts transformers to extract global information at multi-scale by encoding tokenized image patches. The MLiRA-Net establishes local–global spatial features of different scales, and achieves the re-direction of shallow features for upsampling recovery through skip-connections. To evaluate MLiRA-Net, we conduct extensive experiments on the anatomical tracings of lesions after stroke (ATLAS) dataset, and select dice similarity coefficient (DSC) and hausdorff distance (HD) as the major evaluation metrics. Experimental results show that the proposed method has the segmentation performance of 61.19% DSC and 13.49 mm HD. Compared with the existing TransUNet benchmark method, DSC and HD are improved by 4.96% and 5.95 mm, respectively.

LNFCOS: Efficient Object Detection through Deep Learning Based on LNblock

In recent deep-learning-based real-time object detection methods, the trade-off between accuracy and computational cost is an important consideration. Therefore, based on the fully convolutional one-stage detector (FCOS), which is a one-stage object detection method, we propose a light next FCOS (LNFCOS) that achieves an optimal trade-off between computational cost and accuracy. In LNFCOS, the loss of low- and high-level information is minimized by combining the features of different scales through the proposed feature fusion module. Moreover, the light next block (LNblock) is proposed for efficient feature extraction. LNblock performs feature extraction with a low computational cost compared with standard convolutions, through sequential operation on a small amount of spatial and channel information. To define the optimal parameters of LNFCOS suggested through experiments and for a fair comparison, experiments and evaluations were conducted on the publicly available benchmark datasets MSCOCO and PASCAL VOC. Additionally, the average precision (AP) was used as an evaluation index for quantitative evaluation. LNFCOS achieved an optimal trade-off between computational cost and accuracy by achieving a detection accuracy of 79.3 AP and 37.2 AP on the MS COCO and PASCAL VOC datasets, respectively, with 36% lower computational cost than the FCOS.

An improved YOLOv5 method for large objects detection with multi-scale feature cross-layer fusion network

SSD and YOLOv5 are the one-stage object detector representative algorithms. An improved one-stage object detector based on the YOLOv5 method is proposed in this paper, named Multi-scale Feature Cross-layer Fusion Network (M-FCFN). Firstly, we extract shallow features and deep features from the PANet structure for cross-layer fusion and obtain a feature scale different from 80 × 80, 40 × 40, and 20 × 20 as output. Then, according to the single shot multi-box detector, we propose the different scale features which are obtained by cross-layer fusion for dimension reduction and use it as another output for prediction. Therefore, two completely different feature scales are added as the output. Features of different scales are necessary for detecting objects of different sizes, which can increase the probability of object detection and significantly improve detection accuracy. Finally, aiming at the Autoanchor mechanism proposed by YOLOv5, we propose an EIOU k-means calculation. We have compared the four model structures of S , M , L , and X of YOLOv5 respectively. The problem of missed and false detections for large objects is improved which has better detection results. The experimental results show that our methods achieve 89.1% and 67.8% mAP @0.5 on the PASCAL VOC and MS COCO datasets. Compared with the YOLOv5_S, our methods improve by 4.4% and 1.4% mAP @ [0.5:0.95] on the PASCAL VOC and MS COCO datasets. Compared with the four models of YOLOv5, our methods have better detection accuracy for large objects. It should be more attention that our method on the large-scale mAP @ [0.5:0.95] is 5.4% higher than YOLOv5_S on the MS COCO datasets. • We proposed Multi-scale Feature Cross-layer Fusion Network (M-FCFN). • Two completely different feature scales are added as the output. • We propose an EIOU k-means Autoanchor calculation. • The problem of missed and false detections for large objects is improved. • Our method on the large-scale mAP @[0.5:0.95] is 5.4% higher than YOLOv5_S.

Prediction of Prospecting Target Based on Selective Transfer Network

In recent years, with the integration and development of artificial intelligence technology and geology, traditional geological prospecting has begun to change to intelligent prospecting. Intelligent prospecting mainly uses machine learning technology to predict the prospecting target area by mining the correlation between geological variables and metallogenic characteristics, which usually requires a large amount of data for training. However, there are some problems in the actual research, such as fewer geological sample data and irregular mining features, which affect the accuracy and reliability of intelligent prospecting prediction. Taking the Pangxidong study area in Guangdong Province as an example, this paper proposes a deep learning framework (SKT) for prospecting target prediction based on selective knowledge transfer and carries out intelligent prospecting target prediction research based on geochemical data in Pangxidong. The irregular features of different scales in the mining area are captured by dilation convolution, and the weight parameters of the source network are selectively transferred to different target networks for training, so as to increase the generalization performance of the model. A large number of experimental results show that this method has obvious advantages over other state-of-the-art methods in the prediction of prospecting target areas, and the prediction effect in the samples with mines is greatly improved, which can effectively alleviate the problems of a small number of geological samples and irregular features of mining areas in prospecting prediction.

Effects of Microtopography on Runoff Generation in Plain Farmland: New Insights from an Event-Based Rainfall–Runoff Model

Plain farmland areas without significant topographic slope exhibit microtopographic features of different scales. Quantitative assessment of the effects of microtopography at different scales on runoff generation in typical farmland areas is of great significance for regional water resources management and flood disaster forecasting. The main objective of the study was to develop an event-based rainfall–runoff model based on the layered Green–Ampt model (LGAM) with the consideration of plot-scale microtopographic features in plain farmland areas. Our experimental field, located in Taihu Lake Basin, was classified into three types of topographic subunits (i.e., main field, rill, and ditch) according to the average elevation. To simplify the concentration process for three topographic subunits, the average concentration time method was employed. Here, various experimental scenarios were simulated, including two classical unsteady rainfall events in homogeneous soil, one ponding infiltration experiment, and two typical rainfall–runoff events in the experimental field. We found that the multilayered setting showed higher accuracy than the homogeneous setting for simulating infiltration in the ponding infiltration experiment in the field. The RMSE of simulated ponding water depth reduced from 0.166 cm to 0.035 cm and NSE rose from 0.988 to 0.999. The simulated hydrograph considering microtopography effects proved higher accuracy than that under unified topography assumption. After classifying topography, the RMSE and NSE of simulated hydrographs decreased and increased, respectively. The lower the topographic subunit, the earlier the outflow occurred. At the early stage, the runoff mostly originated from the relatively low topographic subunits. Infiltration-excess regime under saturated condition may initially dominate in the low-lying ditch under intense rainfall, with extremely high runoff coefficient. Concentration process in the plain farmland area was affected by both rainfall intensity and microtopography. The greater the rainfall intensity, the shorter the average concentration time. The concentration velocity under heavy rainfall was four times faster than that under light rainfall. The lower topographic subunit was characterized by shorter concentration pathway and average concentration time. Ditches reduced the peak flow and advanced the time to peak. This quantitative study provides new insights into effects of microtopography on runoff generation in plain farmland area as well as an effective alternative for plot-scale rainfall–runoff modeling.

GCLDNet: Gastric cancer lesion detection network combining level feature aggregation and attention feature fusion.

BackgroundAnalysis of histopathological slices of gastric cancer is the gold standard for diagnosing gastric cancer, while manual identification is time-consuming and highly relies on the experience of pathologists. Artificial intelligence methods, particularly deep learning, can assist pathologists in finding cancerous tissues and realizing automated detection. However, due to the variety of shapes and sizes of gastric cancer lesions, as well as many interfering factors, GCHIs have a high level of complexity and difficulty in accurately finding the lesion region. Traditional deep learning methods cannot effectively extract discriminative features because of their simple decoding method so they cannot detect lesions accurately, and there is less research dedicated to detecting gastric cancer lesions.MethodsWe propose a gastric cancer lesion detection network (GCLDNet). At first, GCLDNet designs a level feature aggregation structure in decoder, which can effectively fuse deep and shallow features of GCHIs. Second, an attention feature fusion module is introduced to accurately locate the lesion area, which merges attention features of different scales and obtains rich discriminative information focusing on lesion. Finally, focal Tversky loss (FTL) is employed as a loss function to depress false-negative predictions and mine difficult samples.ResultsExperimental results on two GCHI datasets of SEED and BOT show that DSCs of the GCLDNet are 0.8265 and 0.8991, ACCs are 0.8827 and 0.8949, JIs are 0.7092 and 0.8182, and PREs are 0.7820 and 0.8763, respectively.ConclusionsExperimental results demonstrate the effectiveness of GCLDNet in the detection of gastric cancer lesions. Compared with other state-of-the-art (SOTA) detection methods, the GCLDNet obtains a more satisfactory performance. This research can provide good auxiliary support for pathologists in clinical diagnosis.

Semantic segmentation method of underwater images based on encoder-decoder architecture.

With the exploration and development of marine resources, deep learning is more and more widely used in underwater image processing. However, the quality of the original underwater images is so low that traditional semantic segmentation methods obtain poor segmentation results, such as blurred target edges, insufficient segmentation accuracy, and poor regional boundary segmentation effects. To solve these problems, this paper proposes a semantic segmentation method for underwater images. Firstly, the image enhancement based on multi-spatial transformation is performed to improve the quality of the original images, which is not common in other advanced semantic segmentation methods. Then, the densely connected hybrid atrous convolution effectively expands the receptive field and slows down the speed of resolution reduction. Next, the cascaded atrous convolutional spatial pyramid pooling module integrates boundary features of different scales to enrich target details. Finally, the context information aggregation decoder fuses the features of the shallow network and the deep network to extract rich contextual information, which greatly reduces information loss. The proposed method was evaluated on RUIE, HabCam UID, and UIEBD. Compared with the state-of-the-art semantic segmentation algorithms, the proposed method has advantages in segmentation integrity, location accuracy, boundary clarity, and detail in subjective perception. On the objective data, the proposed method achieves the highest MIOU of 68.3 and OA of 79.4, and it has a low resource consumption. Besides, the ablation experiment also verifies the effectiveness of our method.

MSAA-Net: a multi-scale attention-aware U-Net is used to segment the liver

Automatic segmentation of the liver from CT images is a very challenging task because the shape of the liver in the abdominal cavity varies from person to person and it also often fits closely with other organs. In recent years, with the continuous development of deep learning and the proposal of CNN, the neural network-based segmentation models have shown good performance in the field of image segmentation. Among the many network models, U-Net stands out in the task of medical image segmentation. In this paper, we propose a segmentation network MSAA-Net combining multi-scale features and an improved attention-aware U-Net. We extracted features of different scales on a single feature layer and performed attention perception in the channel dimension. We demonstrate that this architecture improves the performance of U-Net, while significantly reducing computational costs. To address the problem that U-Net’s skip connection is difficult to optimize for merging objects of different sizes, we designed a multi-scale attention gate structure (MAG), which allows the model to automatically learn to focus on targets of different sizes. In addition, MAG can be extended to all structures which contain skip connections, such as U-Net and FCN variants. Our structure was extensively evaluated on the 3Dircadb dataset, and the DICE similarity coefficient of the method for the liver segmentation task was 94.42%, with a much smaller number of model parameters than other attentional models. The experimental results show that MSAA-Net achieves very competitive performance in liver segmentation.

Super-resolved synthetic aperture radar image reconstruction based on multiresolution fusion discrimination

Generative adversarial networks (GANs) are utilized for synthetic aperture radar (SAR) image super-resolution reconstruction, affording realistic texture details. However, existing GANs only discriminate the final generated high-resolution (HR) image after two consecutive upsampling processes, which ignore some high-frequency information of the reconstructed images. To resolve this issue, a multiresolution fusion discrimination (MRFD) algorithm is proposed to discriminate the reconstructed feature maps after each upsampling. First, a multiresolution discrimination process discriminates the authenticity of each upsampled feature map separately, which reduces the image distortion imposed during two consecutive upsampling processes. Besides, multiresolution feature fusion further preserves the consistent high-frequency texture structures. Finally, a multiscale dense network extracts image features in different scales, with multiscale dense block’s dense connections improving parameter utilization. The experimental results on a SAR dataset demonstrate that the proposed MRFD algorithm performs better in reconstructing the texture details of HR images.