Multi-scale Contextual Information Research Articles

Multi-scale organs image segmentation method improved by squeeze-and-attention based on partially supervised learning

Radiotherapy is one of most treatments for tumors. To accurately control the radiation dose distribution and lessen the radiation damage to normal tissues and organs in radiotherapy, it is essential to delineate organs at risk (OARs) precisely. However, manual delineating and some traditional methods are labor-intensive and time-consuming. There is an urgent need for fast and precise segmentation methods in radiotherapy. This paper proposes a fully automatic segmentation method based on the 3D U-Net for multi-organ in head and neck. It introduces squeeze-and-attention blocks to gather multi-scale context information and the receptive field block to balance the performance between large-sized and small-sized organs. Furthermore, it is trained by the marginal and exclusion loss function in a partially supervised learning mode. We evaluated the model with dice similarity coefficient (DSC), 95% Hausdorff distance (95HD) and inference time. Its average DSC is 0.829, which is 4.5%, 3.2%, and 2.4% higher than AnatomyNet's, nnU-net's, and FocusNet's, respectively, and its average 95HD is 2.19. Moreover, its inference time and parameters are 63% and 60% less than FocusNetv2's. For the segmentation of OARs in head and neck, our model is more accurate than AnatomyNet, faster than FocusNetv2, and better balances between segmentation accuracy and inference time. It demonstrates that our method is more applicable for clinical treatment.

ECDNet: A bilateral lightweight cloud detection network for remote sensing images

Cloud detection is one of the critical tasks in remote sensing image pre-processing and it has attracted extensive research interest. In recent years, deep neural networks based cloud detection methods have surpassed the traditional methods (threshold-based methods and conventional machine learning-based methods). However, current approaches mainly focus on improving detection accuracy. The computation complexity and large model size are ignored. To tackle this problem, we propose a lightweight deep learning cloud detection model: Efficient Cloud Detection Network (ECDNet). This model is based on the encoder-decoder structure. In the encoder, a two-path architecture is proposed to extract the spatial and semantic information concurrently. One pathway is the detail branch. It is designed to capture low-level detail spatial features with only a few parameters. The other pathway is the semantic branch, which is mainly for capturing context features. In the semantic branch, a proposed dense pyramid module (DPM) is designed for multi-scale contextual information extraction. The number of parameters and calculations in DPM is greatly reduced by features reusing. Besides, a FusionBlock is developed to merge these two kinds of information. Then the extreme lightweight decoder recovers the cloud mask to the same scale as the input image step by step. To improve performance, boost loss is introduced without inference cost increment. We evaluate the proposed method on two public datasets: LandSat8 and MODIS. Extensive experiments demonstrate that the proposed ECDNet achieves comparable accuracy as the state-of-art cloud detection methods, and meantime has a much smaller model size and less computation burden.

SAR-DRDNet: A SAR image despeckling network with detail recovery

Synthetic aperture radar (SAR) image is affected by inherent speckle, greatly hindering its interpretation and subsequent applications. In SAR images despeckling tasks, it is still challenging to remove speckle while preserving spatial details. Traditional methods generally require a complicated design of relationships between the speckled and noise-free SAR images. The growing popularity of deep learning algorithms exhibits great potential in resolving this limitation. However, the existing deep learning-based despeckling methods need further improvement in speckle suppression and details preservation. Therefore, we propose an end-to-end SAR images despeckling residual network with detail recovery, named SAR-DRDNet, which is mainly composed of non-local blocks (NLBs) and detail recovery blocks (DRBs). NLB takes full advantage of the global information of the SAR image to suppress speckle. DRB is further employed to recover lost details in the process of NLBs, taking into account the multi-scale contextual information of pixels and has a larger receptive field without reducing the resolution of feature maps. We validate the proposed SAR-DRDNet on the simulated and real SAR data. Both quantitative and qualitative comparisons demonstrate the superiority of the proposed SAR-DRDNet over the many mainstream methods, evidenced by its capability to achieve a great balance between speckle suppression and textural details preservation.

MDA-Unet: A Multi-Scale Dilated Attention U-Net for Medical Image Segmentation

The advanced development of deep learning methods has recently made significant improvements in medical image segmentation. Encoder–decoder networks, such as U-Net, have addressed some of the challenges in medical image segmentation with an outstanding performance, which has promoted them to be the most dominating deep learning architecture in this domain. Despite their outstanding performance, we argue that they still lack some aspects. First, there is incompatibility in U-Net’s skip connection between the encoder and decoder features due to the semantic gap between low-processed encoder features and highly processed decoder features, which adversely affects the final prediction. Second, it lacks capturing multi-scale context information and ignores the contribution of all semantic information through the segmentation process. Therefore, we propose a model named MDA-Unet, a novel multi-scale deep learning segmentation model. MDA-Unet improves upon U-Net and enhances its performance in segmenting medical images with variability in the shape and size of the region of interest. The model is integrated with a multi-scale spatial attention module, where spatial attention maps are derived from a hybrid hierarchical dilated convolution module that captures multi-scale context information. To ease the training process and reduce the gradient vanishing problem, residual blocks are deployed instead of the basic U-net blocks. Through a channel attention mechanism, the high-level decoder features are used to guide the low-level encoder features to promote the selection of meaningful context information, thus ensuring effective fusion. We evaluated our model on 2 different datasets: a lung dataset of 2628 axial CT images and an echocardiographic dataset of 2000 images, each with its own challenges. Our model has achieved a significant gain in performance with a slight increase in the number of trainable parameters in comparison with the basic U-Net model, providing a dice score of 98.3% on the lung dataset and 96.7% on the echocardiographic dataset, where the basic U-Net has achieved 94.2% on the lung dataset and 93.9% on the echocardiographic dataset.

Atrous Pyramid GAN Segmentation Network for Fish Images with High Performance

With the development of computer science technology, theory and method of image segmentation are widely used in fish discrimination, which plays an important role in improving the efficiency of fisheries sorting and biodiversity studying. However, the existing methods of fish images segmentation are less accurate and inefficient, which is worthy of in-depth exploration. Therefore, this paper proposes an atrous pyramid GAN segmentation network aimed at increasing accuracy and efficiency. This paper introduces an atrous pyramid structure, and the GAN module is added before the CNN backbone in order to augment the dataset. The Atrous pyramid structure first fuses the input and output of the dilated convolutional layer with a small sampling rate and then feeds the fused features into the subsequent dilated convolutional layer with a large sampling rate to obtain dense multiscale contextual information. Thus, by capturing richer contextual information, this structure improves the accuracy of segmentation results. In addition to the aforementioned innovation, various data enhancement methods, such as MixUp, Mosaic, CutMix, and CutOut, are used in this paper to enhance the model’s robustness. This paper also improves the loss function and uses the label smoothing method to prevent model overfitting. The improvement is also tested by extensive ablation experiments. As a result, our model’s F1-score, GA, and MIoU were tested on the validation dataset, reaching 0.961, 0.981, and 0.973, respectively. This experimental result demonstrates that the proposed model outperforms all the other contrast models. Moreover, in order to accelerate the deployment of the encapsulated model on hardware, this paper optimizes the execution time of the matrix multiplication method on Hbird E203 based on Strassen’s algorithm to ensure the efficient operation of the model on this hardware platform.

Edge Guided Context Aggregation Network for Semantic Segmentation of Remote Sensing Imagery

Semantic segmentation of remote sensing imagery (RSI) has obtained great success with the development of deep convolutional neural networks (DCNNs). However, most of the existing algorithms focus on designing end-to-end DCNNs, but neglecting to consider the difficulty of segmentation in imbalance categories, especially for minority categories in RSI, which limits the performance of RSI semantic segmentation. In this paper, a novel edge guided context aggregation network (EGCAN) is proposed for the semantic segmentation of RSI. The Unet is employed as backbone. Meanwhile, an edge guided context aggregation branch and minority categories extraction branch are designed for a comprehensive enhancement of semantic modeling. Specifically, the edge guided context aggregation branch is proposed to promote entire semantic comprehension of RSI and further emphasize the representation of edge information, which consists of three modules: edge extraction module (EEM), dual expectation maximization attention module (DEMA), and edge guided module (EGM). EEM is created primarily for accurate edge tracking. According to that, DEMA aggregates global contextual features with different scales and the edge features along spatial and channel dimensions. Subsequently, EGM cascades the aggregated features into the decoder process to capture long-range dependencies and further emphasize the error-prone pixels in the edge region to acquire better semantic labels. Besides this, the exploited minority categories extraction branch is presented to acquire rich multi-scale contextual information through an elaborate hybrid spatial pyramid pooling module (HSPP) to distinguish categories taking a small percentage and background. On the Tianzhi Cup dataset, the proposed algorithm EGCAN achieved an overall accuracy of 84.1% and an average cross-merge ratio of 68.1%, with an accuracy improvement of 0.4% and 1.3% respectively compared to the classical Deeplabv3+ model. Extensive experimental results on the dataset released in ISPRS Vaihingen and Potsdam benchmarks also demonstrate the effectiveness of the proposed EGCAN over other state-of-the-art approaches.

EANet: Iterative edge attention network for medical image segmentation

Accurate and automatic segmentation of medical images can greatly assist the clinical diagnosis and analysis. However, it remains a challenging task due to (1) the diversity of scale in the medical image targets and (2) the complex context environments of medical images, including ambiguity of structural boundaries, complexity of shapes, and the heterogeneity of textures. To comprehensively tackle these challenges, we propose a novel and effective iterative edge attention network (EANet) for medical image segmentation with steps as follows. First, we propose a dynamic scale-aware context (DSC) module, which dynamically adjusts the receptive fields to extract multi-scale contextual information efficiently. Second, an edge-attention preservation (EAP) module is employed to effectively remove noise and help the edge stream focus on processing only the boundary-related information. Finally, a multi-level pairwise regression (MPR) module is designed to combine the complementary edge and region information for refining the ambiguous structure. This iterative optimization helps to learn better representations and more accurate saliency maps. Extensive experimental results demonstrate that the proposed network achieves superior segmentation performance to state-of-the-art methods in four different challenging medical segmentation tasks, including lung nodule segmentation, COVID-19 infection segmentation, lung segmentation, and thyroid nodule segmentation. The source code of our method is available at https://github.com/DLWK/EANet

Attention Mechanism Guided Deep Regression Model for Acne Severity Grading

Acne vulgaris is the common form of acne that primarily affects adolescents, characterised by an eruption of inflammatory and/or non-inflammatory skin lesions. Accurate evaluation and severity grading of acne play a significant role in precise treatment for patients. Manual acne examination is typically conducted by dermatologists through visual inspection of the patient skin and counting the number of acne lesions. However, this task costs time and requires excessive effort by dermatologists. This paper presents automated acne counting and severity grading method from facial images. To this end, we develop a multi-scale dilated fully convolutional regressor for density map generation integrated with an attention mechanism. The proposed fully convolutional regressor module adapts UNet with dilated convolution filters to systematically aggregate multi-scale contextual information for density maps generation. We incorporate an attention mechanism represented by prior knowledge of bounding boxes generated by Faster R-CNN into the regressor model. This attention mechanism guides the regressor model on where to look for the acne lesions by locating the most salient features related to the understudied acne lesions, therefore improving its robustness to diverse facial acne lesion distributions in sparse and dense regions. Finally, integrating over the generated density maps yields the count of acne lesions within an image, and subsequently the acne count indicates the level of acne severity. The obtained results demonstrate improved performance compared to the state-of-the-art methods in terms of regression and classification metrics. The developed computer-based diagnosis tool would greatly benefit and support automated acne lesion severity grading, significantly reducing the manual assessment and evaluation workload.

FAFFENet: frequency attention and feature fusion enhancement network for multiscale remote sensing target detection

As one of the most important research directions in the field of computer vision, target detection has made great progress. However, in the field of remote sensing image processing, target detection still has the problem of insufficient feature extraction about the small targets and the low accuracy caused by large-scale difference. We propose a multiscale remote sensing image target detection method based on frequency attention and feature fusion enhancement to achieve better detection results. This method can effectively learn the features extracted by multifrequency components through the frequency channel attention network, realize the interaction of diversified information from various channels, and improve the representation ability of the network. A unique feature pyramid network is designed using interactive up-and-down sampling and skip connection, which sufficiently fuses the features, effectively captures multiscale context information, and improves the detection performance of the model. A feature enhancement network combining dilated convolution and residuals is designed to mine deeper semantic information. The experimental results show that the proposed method achieves higher detection accuracy on the two datasets NWPU VHR-10 and RSOD, when compared with some state-of-the-art target detection methods, and effectively improves the detection performance.

Lesion Segmentation in Gastroscopic Images Using Generative Adversarial Networks.

The segmentation of the lesion region in gastroscopic images is highly important for the detection and treatment of early gastric cancer. This paper proposes a novel approach for gastric lesion segmentation by using generative adversarial training. First, a segmentation network is designed to generate accurate segmentation masks for gastric lesions. The proposed segmentation network adds residual blocks to the encoding and decoding path of U-Net. The cascaded dilated convolution is also added at the bottleneck of U-Net. The residual connection promotes information propagation, while dilated convolution integrates multi-scale context information. Meanwhile, a discriminator is used to distinguish the generated and real segmentation masks. The proposed discriminator is a Markov discriminator (Patch-GAN), which discriminates each [Formula: see text] matrix in the image. In the process of network training, the adversary training mechanism is used to iteratively optimize the generator and the discriminator until they converge at the same time. The experimental results show that the dice, accuracy, and recall are 86.6%, 91.9%, and 87.3%, respectively. These metrics are significantly better than the existing models, which proves the effectiveness of this method and can meet the needs of clinical diagnosis and treatment.

A Pavement Crack Detection Method Based on Multiscale Attention and HFS.

To solve the problem of low detection accuracy due to the loss of detailed information when extracting pavement crack features in traditional U-shaped networks, a pavement crack detection method based on multiscale attention and hesitant fuzzy set (HFS) is proposed. First, the encoding-decoding structure is used to construct a pavement crack segmentation network, ResNeXt50 is used to extract features in the encoding stage, and a multiscale feature fusion module (MFF) is designed to obtain multiscale context information. Second, in the decoding stage, a high-efficiency dual attention module (EDA) is used to enhance the ability of capturing details of the cracks while suppressing background noise. Finally, the membership degree of the crack is calculated based on the advantages of the HFS in multiattribute decision-making to obtain the similarity of the crack, and the binary image after segmentation is judged by the hesitation fuzzy measure. The experiment was conducted on the public road crack dataset Crack500. In terms of segmentation performance, the evaluation indexes Intersection over Union (IoU), Precision, and Dice coefficients of the proposed network reached 55.56%, 74.26%, and 67.43%, respectively; in terms of classification performance, for transversal and longitudinal cracks, the classification accuracy was 84% ± 0.5%, while the block and the alligator were both 78% ± 0.5%. The experimental results prove that the crack details detected by the proposed method are more abundant, and the image detection effect of complex topological structures and small cracks are better.

Multi‐scale attention encoder for street‐to‐aerial image geo‐localization

AbstractThe goal of street‐to‐aerial cross‐view image geo‐localization is to determine the location of the query street‐view image by retrieving the aerial‐view image from the same place. The drastic viewpoint and appearance gap between the aerial‐view and the street‐view images brings a huge challenge against this task. In this paper, we propose a novel multiscale attention encoder to capture the multiscale contextual information of the aerial/street‐view images. To bridge the domain gap between these two view images, we first use an inverse polar transform to make the street‐view images approximately aligned with the aerial‐view images. Then, the explored multiscale attention encoder is applied to convert the image into feature representation with the guidance of the learnt multiscale information. Finally, we propose a novel global mining strategy to enable the network to pay more attention to hard negative exemplars. Experiments on standard benchmark datasets show that our approach obtains 81.39% top‐1 recall rate on the CVUSA dataset and 71.52% on the CVACT dataset, achieving the state‐of‐the‐art performance and outperforming most of the existing methods significantly.

B-FGC-Net: A Building Extraction Network from High Resolution Remote Sensing Imagery

Deep learning (DL) shows remarkable performance in extracting buildings from high resolution remote sensing images. However, how to improve the performance of DL based methods, especially the perception of spatial information, is worth further study. For this purpose, we proposed a building extraction network with feature highlighting, global awareness, and cross level information fusion (B-FGC-Net). The residual learning and spatial attention unit are introduced in the encoder of the B-FGC-Net, which simplifies the training of deep convolutional neural networks and highlights the spatial information representation of features. The global feature information awareness module is added to capture multiscale contextual information and integrate the global semantic information. The cross level feature recalibration module is used to bridge the semantic gap between low and high level features to complete the effective fusion of cross level information. The performance of the proposed method was tested on two public building datasets and compared with classical methods, such as UNet, LinkNet, and SegNet. Experimental results demonstrate that B-FGC-Net exhibits improved profitability of accurate extraction and information integration for both small and large scale buildings. The IoU scores of B-FGC-Net on WHU and INRIA Building datasets are 90.04% and 79.31%, respectively. B-FGC-Net is an effective and recommended method for extracting buildings from high resolution remote sensing images.

Leveraging uncertainty estimation and spatial pyramid pooling for extracting hydrological features from scanned historical topographic maps

ABSTRACT Historical maps are almost the exclusive source to trace back the characteristics of earth before modern earth observation techniques came into being. Processing historical maps is challenging due to the factors such as diverse designs and scales, or inherent noise from painting, aging, and scanning. Our paper is the first to leverage uncertainty estimation under the framework of Bayesian deep learning to model noise inherent in maps for semantic segmentation of hydrological features from scanned topographic historical maps. To distinguish different features with similar symbolization, we integrate atrous spatial pyramid pooling (ASPP) to incorporate multi-scale contextual information. In total, our algorithm yields predictions with an average dice coefficient of 0.827, improving the performance of a simple U-Net by 26%. Our algorithm outputs intuitively interpretable pixel-wise uncertainty maps that capture uncertainty in object boundaries, noise from drawing, aging, and scanning, as well as out-of-distribution designs. We can use the predicted uncertainty potentially to refine segmentation results, locate rare designs, and select reliable features for future GIS analyses.

SDMNet: A Deep-Supervised Dual Discriminative Metric Network for Change Detection in High-Resolution Remote Sensing Images

High-resolution remote sensing image change detection (CD) is one of the main methods to analyze land surface changes. How to effectively distinguish interesting changes and pseudochanges in high-resolution remote sensing images and form accurate and robust CD results is crucial. To deal with these problems, in this letter, we propose a deep-supervised dual discriminative metric network (SDMNet) trained end-to-end for CD in high-resolution bitemporal remote sensing images. A discriminative decoder network is designed in SDMNet to aggregate global and multiscale contextual information, utilizing high stage features to guide the selection of low stage features stage-by-stage to obtain more consistent and robust features. A discriminative implicit metric module is designed in SDMNet to measure the distance between features to detect changes and utilize batch-balanced contrastive loss (BCL) to enlarge the distance difference between unchanged pairs and changed pairs, while alleviate the problem of sample imbalance, and multiple change graph losses are introduced in the intermediate layer of the network for deep supervision. Quantitative evaluation of our method on the CD datasets SYSU, DSIFN, and CLCD demonstrates that the proposed method can provide superior performance than the other state-of-the-art methods.

Unsupervised Balanced Hash Codes Learning With Multichannel Feature Fusion

Unsupervised hashingalgorithms are widely used in large-scale <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">remote sensing images</i> (RSIs) retrieval task. However, existing RSI retrieval algorithms fail to capture the multichannel characteristic of multispectral RSIs and the balanced property of hash codes, which lead the poor performance of RSI retrieval. To tackle these issues, we develop an unsupervised hashing algorithm, namely, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">variational autoencoder balanced hashing</i> (VABH), to leverage multichannel feature fusion and multiscale context information to perform RSI retrieval task. First, multichannel feature fusion module is designed to extract RSI feature information by leveraging the multichannel properties of multispectral RSI. Second, multiscale learning module is developed to learn the multiscale context information of RSIs. Finally, a novel objective function is designed to capture the discrimination and balanced property of hash codes in the hashing learning process. Comprehensive experiments on diverse benchmark have well demonstrated the reasonableness and effectiveness of the proposed VABH algorithm.

Multiscale Context Aggregation Network for Building Change Detection Using High Resolution Remote Sensing Images

The existing methods of building change detection (CD) using remote sensing (RS) images are still deficient in handling scale variation and class imbalance problems, indicating a decrease in the robustness of small-object detection and pseudo-change information. Thus, a novel building CD framework called the multiscale context aggregation network (MSCANet) is proposed. The high-resolution network is integrated into the feature extracting stage to maintain high-resolution representations throughout the whole process. Then, multiscale context information is aggregated using a scale-aware feature pyramid module (FPM). Recognition performance can be improved from discriminant feature representation learning by using a channel–spatial attention module. Furthermore, a class-balanced loss is proposed to reduce the impact of class imbalance in long-tail datasets. Experimental results from using the LEVIR-CD and SZTAKI AirChange benchmark datasets prove the superiority of the MSCANet over the other baseline methods, with improved maximum F1 scores of 5.28 and 8.47, respectively.

MS$^{2}$A-Net: Multiscale Spectral–Spatial Association Network for Hyperspectral Image Clustering

Remote sensing hyperspectral cameras acquire high spectral-resolution data that reveal valuable composition information on the targets (e.g., for Earth observation and environmental applications). The intrinsic high dimensionality and the lack of sufficient numbers of labeled/training samples prevent efficient processing of hyperspectral images (HSIs). HSI clustering can alleviate these limitations. In this study, we propose a multi-scale spectral-spatial association network (MS <inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> A-Net) to cluster HSIs. The backbone of MS <inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> A-Net is an autoencoder architecture that allows the network to capture the non-linear relation between data points in an unsupervised manner. The network applies a multi-stream approach. One stream extracts spectral information by deploying a spectral association unit. The other stream derives multi-scale contextual and spatial information by employing dilated (atrous) convolutional kernels. The obtained feature representation generated by MS <inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> A-Net is fed into a standard k-means clustering algorithm to produce the final clustering result. Extensive experiments on four HSIs for different types of applications (i.e., geological-, rural-, and urban-mapping) demonstrate the superior performance of MS <inline-formula><tex-math notation="LaTeX">$^{2}$</tex-math></inline-formula> A-Net over the state-of-the-art shallow/deep learning-based clustering approaches in terms of clustering accuracy. The code is available at:&#x00A0;<uri>https://github.com/Kasra2020/MS2A-Net</uri>.

Cascaded Multiscale Structure With Self-Smoothing Atrous Convolution for Semantic Segmentation

Convolutional neural networks (CNNs) have attracted great attention in the semantic segmentation of very-high-resolution (VHR) images of urban areas. However, large-scale variation of objects in the urban areas often makes it difficult to achieve good segmentation accuracy. Atrous convolution and atrous spatial pyramid pooling composed of atrous convolution can alleviate this problem by exploring multiscale contextual information. Unfortunately, atrous convolution causes gridding artifacts, where actual receptive fields are separated unit sets and fail to cover all the receptive fields. To address this problem, in this article, we first propose a self-smoothing atrous convolution (SS-AConv) that intrinsically improves atrous convolution, unlike existing methods. SS-AConv enhances sampling rates with low computational costs by adding several key parameters in the spatial dimension of its filter. Then, in the backbone, the Xception network, all max pooling operations are replaced with SS-AConvs for a large and effective receptive field. Moreover, to extract diverse features at multiple scales, we propose the SS-AConv cascaded multiscale structure (SCMS) by integrating SS-AConvs with different rates and the residual correction scheme (RCS) into a cascaded spatial pyramid. Finally, to extract diverse features at dense multiple scales, the SS-AConv convolutional network (SS-ACNet) is constructed by integrating SCMS into both the encoder and decoder layers of the modified Xception network. Extensive experimental results show that SS-ACNet outperforms some state-of-the-art methods on four open challenge datasets: the ISPRS Vaihingen and Potsdam, Cityscapes, and PASCAL VOC 2012.

CDANet: Contextual Detail-Aware Network for High-Spatial-Resolution Remote-Sensing Imagery Shadow Detection

Shadow detection automatically marks shadow pixels in high-spatial-resolution (HSR) imagery with specific categories based on meaningful colorific features. Accurate shadow mapping is crucial in interpreting images and recovering radiometric information. Recent studies have demonstrated the superiority of deep learning in very-high-resolution satellite imagery shadow detection. Previous methods usually overlap convolutional layers but cause the loss of spatial information. In addition, the scale and shape of shadows vary, and the small and irregular shadows are challenging to detect. In addition, the unbalanced distribution of the foreground and the background causes the common binary cross-entropy loss function to be biased, which seriously affects model training. A contextual detail-aware network (CDANet), a novel framework for extracting accurate and complete shadows, is proposed for shadow detection to remedy these issues. In CDANet, a double branch module is embedded in the encoder–decoder structure to effectively alleviate low-level local information loss during convolution. The contextual semantic fusion connection with the residual dilation module is proposed to provide multiscale contextual information of diverse shadows. A hybrid loss function is designed to retain the detailed information of the tiny shadows, which per-pixel calculates the distribution of shadows and improves the robustness of the model. The performance of the proposed method is validated on two distinct shadow detection datasets, and the proposed CDANet reveals higher portability and robustness than other methods.