Dense Skip Connections Research Articles

A deep learning method for optimizing semantic segmentation accuracy of remote sensing images based on improved UNet

Semantic segmentation of remote sensing imagery (RSI) is critical in many domains due to the diverse landscapes and different sizes of geo-objects that RSI contains, making semantic segmentation challenging. In this paper, a convolutional network, named Adaptive Feature Fusion UNet (AFF-UNet), is proposed to optimize the semantic segmentation performance. The model has three key aspects: (1) dense skip connections architecture and an adaptive feature fusion module that adaptively weighs different levels of feature maps to achieve adaptive feature fusion, (2) a channel attention convolution block that obtains the relationship between different channels using a tailored configuration, and (3) a spatial attention module that obtains the relationship between different positions. AFF-UNet was evaluated on two public RSI datasets and was quantitatively and qualitatively compared with other models. Results from the Potsdam dataset showed that the proposed model achieved an increase of 1.09% over DeepLabv3 + in terms of the average F1 score and a 0.99% improvement in overall accuracy. The visual qualitative results also demonstrated a reduction in confusion of object classes, better performance in segmenting different sizes of object classes, and better object integrity. Therefore, the proposed AFF-UNet model optimizes the accuracy of RSI semantic segmentation.

Ensemble Deep Random Vector Functional Link Neural Network for Regression

Inspired by the ensemble strategy of machine learning, deep random vector functional link (dRVFL), and ensemble dRVFL (edRVFL) has shown state-of-the-art results on different datasets. Our present work first fills the gap of dRVFL and edRVFL work in the field of regression. We test and evaluate the performances of the dRVFLs on regression problems. Subsequently, we propose a novel regularization method boosted factor (BF), two dRVFLs variants edRVFL with skip connection (edRVFL-SC) and edRVFL with random skip connections (edRVFL-RSC) and one strategy ensemble skip connection edRVFL (esc-edRVFL) which show significant improvement over the original dRVFL. The BF is a newly introduced hyperparameter to scale the values of the activated hidden neurons to accommodate the diversity of the data, and it is also able to filter the neurons. edRVFL-SC and edRVFL-RSC are the edRVFL variants with skip connections. In edRVFL-SC, we apply dense skip connections to the edRVFL, which is inspired by the residual architecture in the deep learning area. However, due to the specificity of randomized networks, the simple skip connections are probably leading to the reuse of useless features. To address this problem, we propose a random skip connection-based edRVFL, which can keep the diversity in the latent space. esc-RVFL is an ensemble scheme that utilizes several edRVFL-RSC models trained on the different folds of the training dataset. The esc-edRVFL is identified as the best-performing algorithm through a comprehensive evaluation of 31 UCI datasets.

MUGAN: Thermal Infrared Image Colorization Using Mixed-Skipping UNet and Generative Adversarial Network

Common cameras cannot capture high quality image in the night or extreme weather conditions that without enough light, while thermal infrared(TIR) cameras are not limited in this situation. Hence, TIR imaging technique is widely used in military, surveillance, nighttime traffic and other scenarios. However, TIR images are monochromatic, and the majority of details of such images are lost, which are difficult for human or computer system to analyze. Translating TIR images into visible images is beneficial to subsequent observation or further processing. Though there are some advances to realize the transformation from TIR images to color visible images, edge distortion and semantic confusion remain to be solved. Therefore, we propose a Mixed-Skipping UNet(MS-UNet) based image colorization model joint Generative Adversarial Network, which is denoted by MUGAN. Firstly, the dense skip connections of UNet++ and full-scale skip connections of UNet 3+ are combined to form the MS-UNet, which is regarded as the generator. In addition, we design a feature extraction module in MS-UNet to effectively capture the multi-scale features in source image. Then, a novel attention mechanism module is designed for decoding stage which can help decoder of MS-UNet to focus on the important information. Moreover, we explore the effect of different loss functions in TIR image colorization task, and the loss function with excellent performance is selected to further optimize the training process of the model. Extensive experiments prove the superiority of our method in the task of TIR image colorization. The code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/HangyingLiao/MUGAN</uri> .

Deep learning with visual explanations for leakage defect segmentation of metro shield tunnel

Tunnel lining leakage is a crucial indicator of metro shield tunnels’ safety status. For automatic, rapid and accurate detection of leakages, this paper proposes a deep learning-based approach with enhanced efficiency, accuracy and interpretability. First, a total of forty U-shaped semantic segmentation models are developed by coupling UNet and UNet++ with six types of classification convolutional neural networks. Then, multiple evaluation indices, i.e., accuracy, computational complexity, and model complexity are introduced to determine the optimal leakage detection models which achieve a balance between efficiency and accuracy. Finally, Gradient-weighted Class Activation Mapping Plus Plus (Grad-CAM++) is leveraged to understand the mechanisms behind the ‘black-box’ of deep learning-based models. The experimental results show that UNet with encoder EfficientNetB6 is the most applicable model amongst UNet-based models, which gets 84.97 % intersection over union (IoU), requires 115.77 billion floating-point operations (FLOPs), and has 42.49 million parameters. UNet++ with encoder EfficientNetB5 achieves 87.70 % IoU, needs 327.26 billion FLOPs, and has only 30.86 million parameters, thereby becoming the best-performing model among UNet++-based models. The quantitative comparison with existing dominant approaches also proves that our proposed models have high accuracy and low time and space complexities. Furthermore, based on the visual explanations, it concludes that deepening and widening the encoder networks can boost models’ performance, and the two feature fusion methods, namely simple skip connection and dense skip connections, play a crucial role in achieving precise tunnel leakage segmentation. The developed algorithms of this study provide a pixel-wise segmentation basis for fast and accurate quantitative assessment of metro tunnel health.

PDRF-Net: a progressive dense residual fusion network for COVID-19 lung CT image segmentation.

The lungs of patients with COVID-19 exhibit distinctive lesion features in chest CT images. Fast and accurate segmentation of lesion sites from CT images of patients' lungs is significant for the diagnosis and monitoring of COVID-19 patients. To this end, we propose a progressive dense residual fusion network named PDRF-Net for COVID-19 lung CT segmentation. Dense skip connections are introduced to capture multi-level contextual information and compensate for the feature loss problem in network delivery. The efficient aggregated residual module is designed for the encoding-decoding structure, which combines a visual transformer and the residual block to enable the network to extract richer and minute-detail features from CT images. Furthermore, we introduce a bilateral channel pixel weighted module to progressively fuse the feature maps obtained from multiple branches. The proposed PDRF-Net obtains good segmentation results on two COVID-19 datasets. Its segmentation performance is superior to baseline by 11.6% and 11.1%, and outperforming other comparative mainstream methods. Thus, PDRF-Net serves as an easy-to-train, high-performance deep learning model that can realize effective segmentation of the COVID-19 lung CT images.

MRI brain tumor segmentation using residual Spatial Pyramid Pooling-powered 3D U-Net.

Brain tumor diagnosis has been a lengthy process, and automation of a process such as brain tumor segmentation speeds up the timeline. U-Nets have been a commonly used solution for semantic segmentation, and it uses a downsampling-upsampling approach to segment tumors. U-Nets rely on residual connections to pass information during upsampling; however, an upsampling block only receives information from one downsampling block. This restricts the context and scope of an upsampling block. In this paper, we propose SPP-U-Net where the residual connections are replaced with a combination of Spatial Pyramid Pooling (SPP) and Attention blocks. Here, SPP provides information from various downsampling blocks, which will increase the scope of reconstruction while attention provides the necessary context by incorporating local characteristics with their corresponding global dependencies. Existing literature uses heavy approaches such as the usage of nested and dense skip connections and transformers. These approaches increase the training parameters within the model which therefore increase the training time and complexity of the model. The proposed approach on the other hand attains comparable results to existing literature without changing the number of trainable parameters over larger dimensions such as 160 × 192 × 192. All in all, the proposed model scores an average dice score of 0.883 and a Hausdorff distance of 7.84 on Brats 2021 cross validation.

CIFNet: context information fusion network for cloud and cloud shadow detection in optical remote sensing imagery

Accurately and automatically detecting cloud and cloud shadow is one of the key steps in the analysis of optical remote sensing imagery. Currently, most cloud and cloud shadow detection methods are prone to false detection, and some clouds and cloud shadows may be missing from the detection results. Due to the lack of contextual information extraction and fusion capabilities, the accuracy of these cloud detection algorithms cannot be guaranteed. This article proposes a deep learning-based convolutional neural network context information fusion network (named CIFNet) to solve the problem of cloud and cloud shadow detection on Gaofen-1 wide field of view optical remote sensing imageries. First, we preprocess the dataset and make the corresponding labels. Next, we design a Resblock-cloud to capture global and local features and prevent the network from degrading. Then, we utilize a global context fusion module to fuse different levels of global context information by dense skip-connections. Finally, a multiscale context fusion module is designed to extract multiscale contextual relations between cloud and cloud shadow. Experimental results show that the proposed CIFNet method can obtain better cloud and cloud shadow boundaries in complex situations and therefore outperforms the competing methods.

MCA-UNet: multi-scale cross co-attentional U-Net for automatic medical image segmentation.

Medical image segmentation is a challenging task due to the high variation in shape, size and position of infections or lesions in medical images. It is necessary to construct multi-scale representations to capture image contents from different scales. However, it is still challenging for U-Net with a simple skip connection to model the global multi-scale context. To overcome it, we proposed a dense skip-connection with cross co-attention in U-Net to solve the semantic gaps for an accurate automatic medical image segmentation. We name our method MCA-UNet, which enjoys two benefits: (1) it has a strong ability to model the multi-scale features, and (2) it jointly explores the spatial and channel attentions. The experimental results on the COVID-19 and IDRiD datasets suggest that our MCA-UNet produces more precise segmentation performance for the consolidation, ground-glass opacity (GGO), microaneurysms (MA) and hard exudates (EX). The source code of this work will be released via https://github.com/McGregorWwww/MCA-UNet/.

JNMR: Joint Non-linear Motion Regression for Video Frame Interpolation.

Video frame interpolation (VFI) aims to generate predictive frames by motion-warping from bidirectional references. Most examples of VFI utilize spatiotemporal semantic information to realize motion estimation and interpolation. However, due to variable acceleration, irregular movement trajectories, and camera movement in real-world cases, they can not be sufficient to deal with non-linear middle frame estimation. In this paper, we present a reformulation of the VFI as a joint non-linear motion regression (JNMR) strategy to model the complicated inter-frame motions. Specifically, the motion trajectory between the target frame and multiple reference frames is regressed by a temporal concatenation of multi-stage quadratic models. Then, a comprehensive joint distribution is constructed to connect all temporal motions. Moreover, to reserve more contextual details for joint regression, the feature learning network is devised to explore clarified feature expressions with dense skip-connection. Later, a coarse-to-fine synthesis enhancement module is utilized to learn visual dynamics at different resolutions with multi-scale textures. The experimental VFI results show the effectiveness and significant improvement of joint motion regression over the state-of-the-art methods.

Modified UNet with attention gate and dense skip connection for flow field information prediction with porous media

Porous media flow phenomena are commonly found in nature and industry. Understanding the patterns can improve industrial production and efficiency. Traditional numerical methods can predict flow field information within porous media exactly, even on pore scale. However, once the pore structures of porous media are changed, the traditional numerical methods have to start over again, which implies poor extension and low efficiency for realistic application. To address the challenge, we propose a new deep learning method for image segmentation based on convolutional neural networks that can quickly predict the saturation variation and flow velocity profile of multiphase flow directly from the structure of porous media. In the present method, the UNet network structure is improved by replacing parallel splicing with different levels of dense skip connection to extract flow information at different depths. When processing flow topological information, the introduction of attention mechanism allows effective focus on adjusting the flow field edges and front shapes to improve accuracy. Depending on the pore structures information, the saturation of the porous media also can be evaluated. This result shows that the flow field information (saturation and velocity) of porous media can be quickly predicted by deep learning methods. Compared with the computational fluid dynamics (CFD) method, this approach has a better prediction efficiency for porous media. The machine learning method is a good prediction tool for porous media with different porosities and pore structures.

Teeth U-Net: A segmentation model of dental panoramic X-ray images for context semantics and contrast enhancement

Background and objectiveIt is very significant in orthodontics and restorative dentistry that the teeth are segmented from dental panoramic X-ray images. Nevertheless, there are some problems in panoramic X-ray images of teeth, such as blurred interdental boundaries, low contrast between teeth and alveolar bone. MethodsIn this paper, The Teeth U-Net model is proposed in this paper to resolve these problems. This paper makes the following contributions: Firstly, a Squeeze-Excitation Module is utilized in the encoder and the decoder. And proposing a dense skip connection between encoder and decoder to reduce the semantic gap. Secondly, due to the irregular shape of the teeth and the low contrast of the dental panoramic X-ray images. A Multi-scale Aggregation attention Block (MAB) in the bottleneck layer is designed to resolve this problem, which can effectively extract teeth shape features and fuse multi-scale features adaptively. Thirdly, in order to capture dental feature information in a larger field of perception, this paper designs a Dilated Hybrid self-Attentive Block (DHAB) at the bottleneck layer. This module effectively suppresses the task-irrelevant background region information without increasing the network parameters. Finally, the effectiveness of the algorithm is validated using a clinical dental panoramic X-ray image datasets. ResultsThe results of the three comparison experiments are shown that Accuracy, Precision, Recall, Dice, Volumetric Overlap Error and Relative Volume Difference for dental panoramic X-ray teeth segmentation are 98.53%, 95.62%, 94.51%, 94.28%, 88.92% and 95.97% by the proposed model respectively. ConclusionThe proposed modules complement each other in processing every detail of the dental panoramic X-ray images, which can effectively improve the efficiency of preoperative preparation and postoperative evaluation, and promote the application of dental panoramic X-ray in medical image segmentation. There are more accuracy about Teeth U-Net than others model in dental panoramic X-ray teeth segmentation. That is very important to clinical doctors to cure in orthodontics and restorative dentistry.

AFSNet: attention-guided full-scale feature aggregation network for high-resolution remote sensing image change detection

ABSTRACT Using change detection technique precisely analyzes remote sensing images, it has a broad range of applications in resource surveys, surveillance systems, and map updating. In recent years, deep learning-based methods have become a focus area owing to their excellent feature extraction and representation ability. The fusion of multi-scale features is the key to improving change detection performance in fully convolutional network-based structural methods, primarily based on an architecture with skip connections or nested and dense skip connections. However, these methods only fuse features on the same scale and lack sufficient information from multiple scales to generate appropriate results. Traditional feature fusion with redundant and unsupervised information leads to poor model fitting. To solve these problems, we proposed a novel attention-guided full-scale feature aggregation network (AFSNet). The proposed method used a Siamese structure as the backbone network to extract features, which were then aggregated using full-scale skip connections, and an attention mechanism to avoid feature redundancy. Finally, to obtain a highly accurate final change map, a multiple side-outputs fusion strategy was used to fuse the change maps at different scales. To check the reliability of AFSNet, we tested it on two public datasets, the LEVIR-CD and SVCD datasets. The F1-Score/IoU scores improved by 0.57%/0.95% and 1.14%/2.07% in the two datasets, respectively, compared to those obtained using the methods that achieved suboptimal values. The results showed that AFSNet outperforms other mainstream methods while maintaining a good balance between computational costs and model parameters.

LiM-Net: Lightweight multi-level multiscale network with deep residual learning for automatic liver segmentation in CT images

Automatic liver segmentation gained significant attention in the medical realm to deal with liver anomalies. Furthermore, due to advancements in medical imaging, data volume is increasing day-to-day, which seeks the demand for automatic liver segmentation techniques to evade the labour-intensive process of liver delineation currently followed by medical experts. The proposed method is based on the deep learning approach. We exploited the multi-level multiscale feature extraction and fusion concept to uplift the liver segmentation outcome. The computationally efficient pre-activated multiscale Res2Net backbone architecture with channel-wise attention (PARCA) block plugged into the Unet++ architecture to extract the multiscale fine-grained features with refinement and dense skip connections used for the multiscale feature fusion from the various stages of the network. As a result, the fine-grained multiscale features and multiscale feature fusion from diverse stages provide rich contextual feature representation that enhances decoder competence. Further, we optimized the network using a custom loss function that handles the class imbalance and focuses on the complicated samples from the dataset.The efficacy of the proposed lightweight model was tested experimentally using the publicly available 3DIRCADb, CHAOS and LiTS CT datasets. The proposed model achieved the DSC of 97.3%, 95.1%, and 96.3% on the 3DIRCADb, CHAOS, and LiTS datasets. Nevertheless, the proposed network is lightweight and has 7.5 million parameters which are less than the classical Unet and Unet++ architecture. Thus, the proposed heuristics uplift the liver segmentation outcome and significantly reduce the parameters and the model's computational complexity.

Multi-Feature Aggregation for Semantic Segmentation of an Urban Scene Point Cloud

With the rapid development of cities, semantic segmentation of urban scenes, as an important and effective imaging method, can accurately obtain the distribution information of typical urban ground features, reflecting the development scale and the level of greenery in the cities. There are some challenging problems in the semantic segmentation of point clouds in urban scenes, including different scales, imbalanced class distribution, and missing data caused by occlusion. Based on the point cloud semantic segmentation network RandLA-Net, we propose the semantic segmentation networks RandLA-Net++ and RandLA-Net3+. The RandLA-Net++ network is a deep fusion of the shallow and deep features of the point clouds, and a series of nested dense skip connections is used between the encoder and decoder. RandLA-Net3+ is based on the multi-scale connection between the encoder and decoder; it also connects internally within the decoder to capture fine-grained details and coarse-grained semantic information at a full scale. We also propose incorporating dilated convolution to increase the receptive field and compare the improvement effect of different loss functions on sample class imbalance. After verification and analysis of our labeled urban scene LiDAR point cloud dataset—called NJSeg-3D—the mIoU of the RandLA-Net++ and RandLA-Net3+ networks is 3.4% and 3.2% higher, respectively, than the benchmark network RandLA-Net.

Automatic lung tumor segmentation from CT images using improved 3D densely connected UNet.

Accurate lung tumor segmentation has great significance in the treatment planning of lung cancer. However, robust lung tumor segmentation becomes challenging due to the heterogeneity of tumors and the similar visual characteristics between tumors and surrounding tissues. Hence, we developed an improved 3D dense connected UNet (I-3D DenseUNet) to segment various lung tumors from CT images. The nested dense skip connection adopted in the I-3D DenseUNet aims to contribute similar feature maps between encoder and decoder sub-networks. The dense connection used in encoder-decoder blocks also encourages feature propagation and reuse. A robust data augmentation strategy was employed to alleviate over-fitting based on a 3D thin plate spline (TPS) algorithm. We evaluated our method on 938 lung tumors from three datasets consisting of 421 tumors from the Cancer Imaging Archive (TCIA), 450 malignant tumors from the Lung Image Database Consortium (LIDC), and 67 tumors from the private dataset. Experiment results showed an excellent Dice similarity coefficients (DSC) of 0.8316 for the TCIA and LIDC and 0.8167 for the private dataset. The proposed method presents a strong ability in lung tumor segmentation, and it has the potential to help radiologists in lung cancer treatment planning. Framework of the proposed lung tumor segmentation method.

R2U++: a multiscale recurrent residual U-Net with dense skip connections for medical image segmentation

U-Net is a widely adopted neural network in the domain of medical image segmentation. Despite its quick embracement by the medical imaging community, its performance suffers on complicated datasets. The problem can be ascribed to its simple feature extracting blocks: encoder/decoder, and the semantic gap between encoder and decoder. Variants of U-Net (such as R2U-Net) have been proposed to address the problem of simple feature extracting blocks by making the network deeper, but it does not deal with the semantic gap problem. On the other hand, another variant UNET++ deals with the semantic gap problem by introducing dense skip connections but has simple feature extraction blocks. To overcome these issues, we propose a new U-Net based medical image segmentation architecture R2U++. In the proposed architecture, the adapted changes from vanilla U-Net are: (1) the plain convolutional backbone is replaced by a deeper recurrent residual convolution block. The increased field of view with these blocks aids in extracting crucial features for segmentation which is proven by improvement in the overall performance of the network. (2) The semantic gap between encoder and decoder is reduced by dense skip pathways. These pathways accumulate features coming from multiple scales and apply concatenation accordingly. The modified architecture has embedded multi-depth models, and an ensemble of outputs taken from varying depths improves the performance on foreground objects appearing at various scales in the images. The performance of R2U++ is evaluated on four distinct medical imaging modalities: electron microscopy, X-rays, fundus, and computed tomography. The average gain achieved in IoU score is 1.5 ± 0.37% and in dice score is 0.9 ± 0.33% over UNET++, whereas, 4.21 ± 2.72 in IoU and 3.47 ± 1.89 in dice score over R2U-Net across different medical imaging segmentation datasets.

UNFusion: A Unified Multi-Scale Densely Connected Network for Infrared and Visible Image Fusion

Infrared image retains typical thermal targets while visible image preserves rich texture details, image fusion aims to reconstruct a synthesized image containing prominent targets and abundant texture details. Most of deep learning-based methods mainly focus on convolution operation to extract the local features, but do not fully consider their multi-scale characteristics and global dependencies, which may cause loss of target regions and texture details in the fused image. Towards this goal, we present a unified multi-scale densely connected fusion network in this paper, named as UNFusion. We carefully design a multi-scale encoder-decoder architecture that can efficiently extract and reconstruct multi-scale deep features. Dense skip connections are employed in both encoder and decoder sub-networks to reuse all the intermediate features of different layers and scales for fusion tasks. In the fusion layer, <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$L_{p} $ </tex-math></inline-formula> normalized attention models, which include three kinds of different norms, are proposed to highlight and combine these deep features from spatial and channel dimensions, and the combined spatial and channel attention maps are used to reconstruct a final fused image. We conduct extensive experiments on the public TNO and Roadscene datasets, and the results demonstrate that our UNFusion can simultaneously preserve high brightness of typical thermal targets and abundant texture details to obtain superior scene representation and better visual perception. Besides, our UNFusion achieves better fusion performance and transcends other state-of-the-art methods in terms of qualitative and quantitative comparisons. Our code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Zhishe-Wang/UNFusion</uri> .

SCU-Net++: A Nested U-Net Based on Sharpening Filter and Channel Attention Mechanism

U-Net++ is one of the most prominent deep convolutional neural networks in the field of medical image segmentation after U-Net. However, the semantic gaps between the encoder and decoder subnets are still large, which will lead to fuzzy feature maps and even target regions of segmentation. To solve this problem, we propose an improved semantic segmentation model utilizing channel attention mechanism and Laplacian sharpening filter, SCU-Net++: dense skip connections are redesigned with sharpening filters to ease the semantic gaps, and channel attention modules are used to make the model pay more attention on the feature maps that are useful for our pixel-level classification task. Compared with U-Net++, the proposed model obtains a more competitive performance on the Pancreas Segmentation dataset and Liver Tumor Segmentation dataset, while increases a very small number of learnable parameters and thus almost does not make additional training and reasoning costs. The training of the proposed method is carried out in deep supervision mode, which alleviates the problem of gradient disappearance, and pruning mechanism can be activated to accelerate the reasoning speed.

A Self-Attentive Hybrid Coding Network for 3D Change Detection in High-Resolution Optical Stereo Images

Real-time monitoring of urban building development provides a basis for urban planning and management. Remote sensing change detection is a key technology for achieving this goal. Intelligent change detection based on deep learning of remote sensing images is a current focus of research. However, most methods only use unimodal remote sensing data and ignore vertical features, leading to incomplete characterization, poor detection of small targets, and false detections and omissions. To solve these problems, we propose a multi-path self-attentive hybrid coding network model (MAHNet) that fuses high-resolution remote sensing images and digital surface models (DSMs) for 3D change detection of urban buildings. We use stereo images from the Gaofen-7 (GF-7) stereo mapping satellite as the data source. In the encoding stage, we propose a multi-path hybrid encoder, which is a structure that can efficiently perform multi-dimensional feature mining of multimodal data. In the deep feature fusion link, a dual self-attentive fusion structure is designed that can improve the deep feature fusion and characterization of multimodal data. In the decoding stage, a dense skip-connection decoder is designed that can fuse multi-scale features flexibly and reduce spatial information losses in small-change regions in the down-sampling process, while enhancing feature utilization and propagation efficiency. Experimental results show that MAHNet achieves accurate pixel-level change detection in complex urban scenes with an overall accuracy of 97.44% and F1-score of 92.59%, thereby outperforming other methods of change detection.

Adversarial Multiscale Feature Learning Framework for Overlapping Chromosome Segmentation.

Chromosome karyotype analysis is of great clinical importance in the diagnosis and treatment of diseases. Since manual analysis is highly time and effort consuming, computer-assisted automatic chromosome karyotype analysis based on images is routinely used to improve the efficiency and accuracy of the analysis. However, the strip-shaped chromosomes easily overlap each other when imaged, significantly affecting the accuracy of the subsequent analysis and hindering the development of chromosome analysis instruments. In this paper, we present an adversarial, multiscale feature learning framework to improve the accuracy and adaptability of overlapping chromosome segmentation. We first adopt the nested U-shaped network with dense skip connections as the generator to explore the optimal representation of the chromosome images by exploiting multiscale features. Then we use the conditional generative adversarial network (cGAN) to generate images similar to the original ones; the training stability of the network is enhanced by applying the least-square GAN objective. Finally, we replace the common cross-entropy loss with the advanced Lovász-Softmax loss to improve the model’s optimization and accelerate the model’s convergence. Comparing with the established algorithms, the performance of our framework is proven superior by using public datasets in eight evaluation criteria, showing its great potential in overlapping chromosome segmentation.