Pixel Classification Research Articles

Reversible data hiding using morphology based pixel classification

Pixel value ordering (PVO) prediction has become a research hotspot in reversible data hiding community and existing achievements have proved the necessity of predicting more pixels in a block. The increase in predicted pixels is beneficial to fully exploiting the redundancy of the cover image, but a single reference pixel may lead to decreased prediction quality. Given this situation, this paper proposes to develop PVO in correlated pixels construction and introduce morphology as an important theoretical tool. Several largest pixels in a block are first selected as candidate reference pixels which are to be activated in descending order. Except for the last one, each active reference pixel is used to predict all pixels to generate a correlation matrix, on which the dilation operation is carried out to pick out connected pixels that are likely to be strongly correlated with the reference pixel. Whether to activate the next reference pixel depends on the presence of disconnected pixels. Once the last reference pixel is activated, all remaining pixels will be taken as connected pixels. After the first round of embedding ends, several smallest pixels are similarly selected as candidate reference pixels to realize the second round of embedding. Experimental results demonstrate that satisfactory superiority in fidelity is achieved, e.g., the average PSNR for the Kodak image database is 63.85 dB with an embedding capacity of 10,000 bits.

CSDFormer: A cloud and shadow detection method for landsat images based on transformer

Cloud and shadow (CS) detection is crucial prerequisite for application of remote sensing images. Current deep learning-based detection algorithms mainly employ Convolutional Neural Networks (CNNs). However, the local receptive field in CNNs cannot effectively capture global contextual information, which hinders accurate characterization of the dependency between clouds and shadows. In vision Transformers, self-attention mechanisms can effectively capture the long-distance dependencies between different regions in an image. Inspired by this, this paper proposed a new CS Detection algorithm based on a Transformer, called CSDFormer. Specifically, we exclusively employed a hierarchical Transformer structure in the encoder stage to extract features of CS. Each Transformer layer contains several multi-head self-attention mechanisms for calculating pixel-wise long-distance connectivity. The designed structure enables the Transformer to better extract global context information, which helps to strengthen the comprehension of the semantic relationships between clouds and shadows. Benefiting from the global feature extraction capability of the encoder stage, we employed several simple multilayer perceptron layers for multi-scale feature map fusion and pixel classification in the decoder stage. The proposed CSDFormer was validated using 898 Landsat 8 Biome images with 512 × 512 pixels, producing an overall accuracy of 95.28 % and a mean intersection over union of 84.08 %, outperforming three state-of-the-art CNN-based algorithms. CSDFormer is consistently more accurate in detection of both clouds and shadows. Owing to the parallel computing capability of the self-attention mechanism, CSDFormer is computationally more efficient than the three CNN-based benchmark methods. For the input spectral bands, the performance of CSDFormer produced can be further enhanced with additional thermal infrared bands.

Spatial Resolution Enhancement of Vegetation Indexes via Fusion of Hyperspectral and Multispectral Satellite Data

The definition and calculation of a spectral index suitable for characterizing vegetated landscapes depend on the number and widths of the bands of the imaging instrument. Here, we point out the advantages of performing the fusion of hyperspectral (HS) satellite data with the multispectral (MS) bands of Sentinel-2 to calculate such vegetation indexes as the normalized area over reflectance curve (NAOC) and the red-edge inflection point (REIP), which benefit from the availability of quasi-continuous pixel spectra. Unfortunately, MS data may be acquired from satellite platforms with very high spatial resolution; HS data may not. Despite their excellent spectral resolution, satellite imaging spectrometers currently resolve areas not greater than 30 × 30 m2, where different thematic classes of landscape may be mixed together to form a unique pixel spectrum. A way to resolve mixed pixels is to perform the fusion of the HS dataset with the same dataset produced by an MS scanner that images the same scene with a finer spatial resolution. The HS dataset is sharpened from 30 m to 10 m by means of the Sentinel-2 bands that have all been previously brought to 10 m. To do so, the hyper-sharpening protocol, that is, m:n fusion, is exploited in two nested steps: the first one to bring the 20 m bands of Sentinel-2 all to 10 m, the second one to sharpen all the 30 m HS bands to 10 m by using the Sentinel-2 bands previously hyper-sharpened to 10 m. Results are presented on an agricultural test site in The Netherlands imaged by Sentinel-2 and by the satellite imaging spectrometer recently launched as a part of the environmental mapping and analysis program (EnMAP). Firstly, the excellent match of statistical consistency of the fused HS data to the original MS and HS data is evaluated by means of analysis tools, existing and developed ad hoc for this specific case. Then, the spatial and radiometric accuracy of REIP and NAOC calculated from fused HS data are analyzed on the classes of pure and mixed pixels. On pure pixels, the values of REIP and NAOC calculated from fused data are consistent with those calculated from the original HS data. Conversely, mixed pixels are spectrally unmixed by the fusion process to resolve the 10 m scale of the MS data. How the proposed method can be used to check the temporal evolution of vegetation indexes when a unique HS image and many MS images are available is the object of a final discussion.

EAIS-Former: An efficient and accurate image segmentation method for fruit leaf diseases

Fruit leaf disease segmentation is an essential foundation for achieving accurate disease diagnosis and identification. However, shadows caused by folded leaves and serrations on leaves can lead to difficulty in extracting edge features, affecting the accuracy of leaf segmentation. In addition, the varying shapes and blurred boundaries of disease spots can further lead to poor segmentation performance of spots. To address the above problems, this work proposes a method called EAIS-Former by combining the advantages of global modeling of Transformer, local processing and positional coding of convolutional neural network (CNN) for accurate segmentation in fruit leaf disease images. Dual scale overlap (DSO) patch embedding is designed to effectively extract multi-scale disease features by dual paths to alleviate omission of lesions. Ultra large convolution (ULC) Transformer block is customized for performing positional encoding and global modeling to efficiently extract global and positional features of leaves and diseases. Skip convolutional local optimization (SCLO) module is proposed to optimize the local detail and edge information and improve the pixel classification ability of the model so that the segmentation results of leaves and spots can be finer and more tiny spots can be extracted. Double layer upsampling (DLU) decoder is built to efficiently fuse the detail information with the semantic information and output the accurate segmentation results of leaves and spots. The experimental results show that the proposed method reach 99.04%, 98.64%, 99.24%, 99.42%, 98.59% and 98.58% intersection over union (IoU) for leaf segmentation on apple rust, pomegranate cercospora spot, mango anthracnose, jamun fungal disease, apple alternaria blotch and apple gray spot datasets, respectively. The IoU of lesion segmentation achieve 94.47%, 94.54%, 83.83%, 86.60%, 89.59% and 88.76%, respectively. In contrast to DeepLabv3+, the accuracy of disease segmentation is raised by 5.25%, 5.15%, 5.55%, 7.64%, 7.04% and 9.35%, respectively. Compared with U-Net, the proposed method improves the accuracy of disease spot segmentation by 4.3%, 4.44%, 5.26%, 9.42%, 5.87% and 6.53% under the six fruit leaf test sets, respectively. In addition, total parameters and FLOPs of the proposed method are only 18.44% and 8.47% of U-Net, respectively. Therefore, this study can provide an efficient and accurate method for the task of fruit leaf disease spot segmentation, which provides a sufficient basis for the accurate analysis of fruit leaves and diseases.

Semantic Segmentation of Surface Cracks in Urban Comprehensive Pipe Galleries Based on Global Attention

Cracks inside urban underground comprehensive pipe galleries are small and their characteristics are not obvious. Due to low lighting and large shadow areas, the differentiation between the cracks and background in an image is low. Most current semantic segmentation methods focus on overall segmentation and have a large perceptual range. However, for urban underground comprehensive pipe gallery crack segmentation tasks, it is difficult to pay attention to the detailed features of local edges to obtain accurate segmentation results. A Global Attention Segmentation Network (GA-SegNet) is proposed in this paper. The GA-SegNet is designed to perform semantic segmentation by incorporating global attention mechanisms. In order to perform precise pixel classification in the image, a residual separable convolution attention model is employed in an encoder to extract features at multiple scales. A global attention upsample model (GAM) is utilized in a decoder to enhance the connection between shallow-level features and deep abstract features, which could increase the attention of the network towards small cracks. By employing a balanced loss function, the contribution of crack pixels is increased while reducing the focus on background pixels in the overall loss. This approach aims to improve the segmentation accuracy of cracks. The comparative experimental results with other classic models show that the GA SegNet model proposed in this study has better segmentation performance and multiple evaluation indicators, and has advantages in segmentation accuracy and efficiency.

Abstract TMP120: Contralesional Corticospinal Tract Reinnervates Cervical Spinal Anatomical Regions With Highest Pre-Lesion Synapse Density After Unilateral M1 Stroke in Mice

Introduction: Stroke leaves many patients with significant motor impairment. Following ischemic lesions of the rodent primary motor cortex, sprouting of the contralesional corticospinal tract (cCST) into the injured hemicord is associated with improved motor recovery. In adult mice, most corticospinal neurons project to spinal interneurons, and few synapse directly on alpha motor neurons. We examined the distribution of newly formed corticospinal synapses in the mouse spinal cord, using genetically encoded tracers, 3D microscopy, and a spinal atlas with machine learning-based automated classification and registration. Hypothesis: We tested the hypothesis that cCST neurons innervate homotopic regions in the injured hemicord after unilateral motor cortex stroke in adult mice. Method: We induced a photothrombotic motor cortex stroke or performed sham surgery (n=4 per group) in 8-11 week old male C57/B6 mice and administered a contralesional motor cortex injection of an anterograde adeno-associated virus expressing both membrane-targeted tdTomato and synaptically-targeted eGFP. Cervical spinal cords were subjected to volumetric imaging via serial two-photon tomography (TissueCyte 1000) to characterize synaptic and axonal density of cCST collaterals in whole cervical spinal cords in both stroke and sham mice. Unbiased, global quantification of axonal and synaptic density across the entire cervical cord was accomplished by the development of a custom automated image analysis pipeline, incorporating a novel 3D spinal cord reference volume, published spinal cord annotations comprising 47 distinct anatomical regions (SpinalJ), and machine learning based pixel classification. Results: We observed cCST synapses in the 6-week post-stroke injured hemicord innervating lamina IV & IX most densely at C7 & C8 (See Figure). Conclusions: Contralesional CST reinnervation of the denervated hemicord is targeted to both lamina and cervical level specific.

Offshore Oil Spill Detection Based on CNN, DBSCAN, and Hyperspectral Imaging.

Offshore oil spills have the potential to inflict substantial ecological damage, underscoring the critical importance of timely offshore oil spill detection and remediation. At present, offshore oil spill detection typically combines hyperspectral imaging with deep learning techniques. While these methodologies have made significant advancements, they prove inadequate in scenarios requiring real-time detection due to limited model detection speeds. To address this challenge, a method for detecting oil spill areas is introduced, combining convolutional neural networks (CNNs) with the DBSCAN clustering algorithm. This method aims to enhance the efficiency of oil spill area detection in real-time scenarios, providing a potential solution to the limitations posed by the intricate structures of existing models. The proposed method includes a pre-feature selection process applied to the spectral data, followed by pixel classification using a convolutional neural network (CNN) model. Subsequently, the DBSCAN algorithm is employed to segment oil spill areas from the classification results. To validate our proposed method, we simulate an offshore oil spill environment in the laboratory, utilizing a hyperspectral sensing device to collect data and create a dataset. We then compare our method with three other models-DRSNet, CNN-Visual Transformer, and GCN-conducting a comprehensive analysis to evaluate the advantages and limitations of each model.

Markedly Enhanced Analysis of Mass Spectrometry Images Using Weakly Supervised Machine Learning.

Supervised and unsupervised machine learning algorithms are routinelyapplied to time-of-flight secondary ion mass spectrometry (ToF-SIMS) imaging data and, more broadly, to mass spectrometry imaging (MSI). These algorithms have accelerated large-scale, single-pixel analysis, classification, and regression. However, there is relatively little research on methods suited for so-called weakly supervised problems, where ground-truth class labels exist at the image level, but not at the individual pixel level. Unsupervised learning methods are usually applied to these problems. However, these methods cannot make use of available labels. Here a novel method specifically designed for weakly supervised MSI data is presented. A dual-stream multiple instance learning (MIL) approach is adapted from computational pathology that reveals the spatial-spectral characteristics distinguishing different classes of MSI images. The method uses an information entropy-regularized attention mechanism to identify characteristic class pixels that are then used to extract characteristic mass spectra. This work provides a proof-of-concept exemplification using printed ink samples imaged by ToF-SIMS. A second application-oriented study is also presented, focusing on the analysis of a mixed powder sample type. Results demonstrate the potential of the MIL method for broader application in MSI, with implications for understanding subtle spatial-spectral characteristics in various applications and contexts.

Improved Loss Function for Mass Segmentation in Mammography Images Using Density and Mass Size.

Mass segmentation is one of the fundamental tasks used when identifying breast cancer due to the comprehensive information it provides, including the location, size, and border of the masses. Despite significant improvement in the performance of the task, certain properties of the data, such as pixel class imbalance and the diverse appearance and sizes of masses, remain challenging. Recently, there has been a surge in articles proposing to address pixel class imbalance through the formulation of the loss function. While demonstrating an enhancement in performance, they mostly fail to address the problem comprehensively. In this paper, we propose a new perspective on the calculation of the loss that enables the binary segmentation loss to incorporate the sample-level information and region-level losses in a hybrid loss setting. We propose two variations of the loss to include mass size and density in the loss calculation. Also, we introduce a single loss variant using the idea of utilizing mass size and density to enhance focal loss. We tested the proposed method on benchmark datasets: CBIS-DDSM and INbreast. Our approach outperformed the baseline and state-of-the-art methods on both datasets.

Two‐dimensional medical image segmentation based on U‐shaped structure

AbstractWith rapid developments in convolutional neural networks for image processing, deep learning methods based on pixel classification have been extensively applied in medical image segmentation. One popular strategy for such tasks is the encoder‐decoder‐based U‐Net architecture and its variants. Most segmentation methods based on fully convolutional networks will cause the loss of spatial and contextual information due to continuous pooling operations or strided convolution when decreasing image resolution, and make less use of contextual information and global information under different receptive fields. To overcome this shortcoming, this paper proposes a novel structure called RAAU‐Net. In our proposed RAAU‐Net structure, which is a modified U‐shaped architecture, we aim to capture high‐level information while preserving spatial information and focusing on the regions of interest. RAAU‐Net comprises three main components: a feature encoder module that utilizes a pre‐trained ResNet‐18 model as a fixed feature extractor, a multi‐receptive field extraction module that we developed, and a feature decoder module. We have tested our method on several 2D medical image segmentation tasks such as retinal nerve, breast tumor, skin lesion, lung, gland, and polyp segmentation. All the indexes of the model reached the best in the dataset of skin lesions, in which Accuracy, Precision, IoU, Recall, and Dice Score were 3.26%, 5.42%, 9.92%, 6.52%, and 5.95% higher than UNet.

Koncepcja oceny korozji powierzchniowej stalowego mostu kolejowego z zastosowaniem aktywnego systemu teledetekcji bliskiego zasięgu

The concept of assessing surface corrosion of a steel railway bridge was born as a response to the industry need for quick and non-manual assessment of the progress of surface damage of monochromatic bridges. The fourth coordinate of the point cloud, through imaging algorithms, can indicate homogeneous surfaces of the bridge, and therefore subject to corrosion processes or not. The intensity of the object's point cloud, through the use of unsupervised classification tools, ensures the detection of changes in the surface properties of a mono-colored railway bridge. The classification method of unsupervised raster representation of grayscale reflectance intensity (generated from TLS data), as in the case of classical remote sensing, produces classes of pixels with similar reflectance properties. The concept of scientific research on the detection of corrosion progress on a steel railway bridge using an active short-range remote sensing system assumed the development of algorithmic advances allowing for the comparison of periodic classifications of rasters from point clouds. Thanks to the differentiation of images, it is possible to determine changes in the location and extent of corrosion, the rate of its progress in the case of aged steel objects, and the detection of scratches and cracks as critical points of the structure, proving the fulfillment of the operational capabilities of the object, as provided for in the technical documentation. The study indicated the empirical basis for conducting research on automatic corrosion detection.

Machine-learning-guided quantitative delineation of cell morphological features and responses to nanomaterials.

Delineation of cell morphological features is essential to decipher cell responses to external stimuli like theranostic nanomaterials. Conventional methods rely on labeled approaches, such as fluorescence imaging and flow cytometry, to assess cell responses. Besides potentially perturbing cell structure and morphology, these approaches are relatively complex, time-consuming, expensive, and may not be compatible with downstream analysis involving live cells. Herein, leveraging label-free phase-contrast or brightfield microscopy imaging and machine learning, the delineation of different cell types, phenotypes, and states for monitoring live cell responses is reported. Notably, pixel classification based on a supervised random forest classifier is used to distinguish between cells and backgrounds from the microscopy images, followed by cell segmentation and morphological feature extraction. Quantitative analysis shows that most of the compared cell groups have distinguishable size and shape features. Principal component analysis and unsupervised k-means clustering of morphological features reveal the possible existence of heterogenous cell subpopulations and treatment responses among the seemingly homogenous cell groups. This shows the merit of the reported approach in complementing conventional techniques for cell analysis. It is anticipated that the demonstrated method will further aid the implementation of machine learning to streamline the analysis of cell morphology and responses for early disease diagnosis and treatment response monitoring.

Satellite-based landslide distribution mapping with the adoption of deep learning approach in the Kuantan River Basin, Pahang

Landslides are one of the major geological phenomena that is widespread across the globe and have caused destructive outcomes to human life and the overall economic system. Tedious work is required to conventionally collect all evidence of multiple sizes of landslide occurrences in such a huge, developing city, including the Kuantan River Basin (KRB). In fact, landslides are difficult to identify in remote areas, such as in thick and mountainous areas, if no aerial devices or sensor technology is provided at the incident area. Ironically, the landslide distribution map is a useful tool that helps in staging the landslide mitigation plan for landslide-prone areas. Thus, the objectives of this study are (i) to identify landslide events using deep learning and vegetation index approaches on optical satellite data; and (ii) to develop landslide distribution mapping in KRB using the best approach. Remotely sensed optical images of Landsat 8 OLI and Worldview-2 were used to map the landslide distribution and study the spectral pattern of the landslide area. Normalized Difference Vegetation Index (NDVI) were generated for two consecutive years, which is from the year 2022 to 2023. Spectral bands in red and infrared are used to generate the NDVI for visual interpretation of landslide occurrences. The deep learning based on Convolutional Neural Network (CNN) model were used for the pixel classification process. The main output of this study would be a landslide distribution map for the KRB area with high accuracy. The result has also been verified using drone monitoring at the incident sites, which was able to improve landslide detection in tropical environments. Landslide distribution maps accuracy was measured by using the ROC-AUC method, the map accuracy is 88.9%. This map should help the government and private sector plan for the city’s urban development and provide proper planning for geohazard mitigation. An accurate landslide distribution map could be a source of reference for the National Disaster Management Authority (NADMA) for a quick rescue during emergency.

Effective Semi-Supervised Medical Image Segmentation with Probabilistic Representations and Prototype Learning.

Label scarcity, class imbalance and data uncertainty are three primary challenges that are commonly encountered in the semi-supervised medical image segmentation. In this work, we focus on the data uncertainty issue that is overlooked by previous literature. To address this issue, we propose a probabilistic prototype-based classifier that introduces uncertainty estimation into the entire pixel classification process, including probabilistic representation formulation, probabilistic pixel-prototype proximity matching, and distribution prototype update, leveraging principles from probability theory. By explicitly modeling data uncertainty at the pixel level, model robustness of our proposed framework to tricky pixels, such as ambiguous boundaries and noises, is greatly enhanced when compared to its deterministic counterpart and other uncertainty-aware strategy. Empirical evaluations on three publicly available datasets that exhibit severe boundary ambiguity show the superiority of our method over several competitors. Moreover, our method also demonstrates a stronger model robustness to simulated noisy data. Code is available at https://github.com/IsYuchenYuan/PPC.

Sentinel-2 image based smallholder crops classification and accuracy assessment by UAV data

Timely and accurate extraction of crop planting units plays a critical role in crop yield estimation, soil management, food supplies, and disaster warnings. However, precisely mapping crop types is challenging in smallholder farming systems due to heterogeneous and mixed pixels, where field sizes are small, and crop types are very diverse. In this paper, the crop type system of Huocheng County in northwest China as an example area, crop classification feature variables are constructed using Sentinel-2 remote sensing images, and combining the ReliefF algorithm, three feature selection classification models are established by 3092 crop type sampling points data. The crop planting units are extracted using pixel-based and object-based classification methods, respectively, and unmanned aerial vehicle (UAV) data assess the accuracy as ground truth. The Sentinel-2 image based smallholder crops classification results indicate that: The effective combination of optimal input feature variables selection and classification models significantly improves crop classification accuracy. This is particularly evident when integrating this approach into pixel-based classification, addressing challenges such as low pixel classification accuracy in regions of this type. The validation results from UAV data also confirm these findings. At validation points 2, 3, and 4, pixel-based classification demonstrates significantly higher accuracy compared to object-based classification, with pixel-based classification accuracy improving by 9.16%, 7.83%, and 31.83%, respectively. Although the UAV validation accuracy does not reach a high level, pixel-based classification remains the optimal choice for smallholder crop classification. This method offers new insights and references for research related to the classification of smallholder crops with complex planting structures.

Multibranch semantic image segmentation model based on edge optimization and category perception.

In semantic image segmentation tasks, most methods fail to fully use the characteristics of different scales and levels but rather directly perform upsampling. This may cause some effective information to be mistaken for redundant information and discarded, which in turn causes object segmentation confusion. As a convolutional layer deepens, the loss of spatial detail information makes the segmentation effect achieved at the object boundary insufficiently accurate. To address the above problems, we propose an edge optimization and category-aware multibranch semantic segmentation network (ECMNet). First, an attention-guided multibranch fusion backbone network is used to connect features with different resolutions in parallel and perform multiscale information interaction to reduce the loss of spatial detail information. Second, a category perception module is used to learn category feature representations and guide the pixel classification process through an attention mechanism to optimize the resulting segmentation accuracy. Finally, an edge optimization module is used to integrate the edge features into the middle and the deep supervision layers of the network through an adaptive algorithm to enhance its ability to express edge features and optimize the edge segmentation effect. The experimental results show that the MIoU value reaches 79.2% on the Cityspaces dataset and 79.6% on the CamVid dataset, that the number of parameters is significantly lower than those of other models, and that the proposed method can effectively achieve improved semantic image segmentation performance and solve the partial category segmentation confusion problem, giving it certain application prospects.

MIS-Net: A deep learning-based multi-class segmentation model for CT images.

The accuracy of traditional CT image segmentation algorithms is hindered by issues such as low contrast and high noise in the images. While numerous scholars have introduced deep learning-based CT image segmentation algorithms, they still face challenges, particularly in achieving high edge accuracy and addressing pixel classification errors. To tackle these issues, this study proposes the MIS-Net (Medical Images Segment Net) model, a deep learning-based approach. The MIS-Net model incorporates multi-scale atrous convolution into the encoding and decoding structure with symmetry, enabling the comprehensive extraction of multi-scale features from CT images. This enhancement aims to improve the accuracy of lung and liver edge segmentation. In the evaluation using the COVID-19 CT Lung and Infection Segmentation dataset, the left and right lung segmentation results demonstrate that MIS-Net achieves a Dice Similarity Coefficient (DSC) of 97.61. Similarly, in the Liver Tumor Segmentation Challenge 2017 public dataset, the DSC of MIS-Net reaches 98.78.

SegHSI: Semantic Segmentation of Hyperspectral Images with Limited Labeled Pixels.

Hyperspectral images (HSIs), with hundreds of narrow spectral bands, are increasingly used for ground object classification in remote sensing. However, many HSI classification models operate pixel-by-pixel, limiting the utilization of spatial information and resulting in increased inference time for the whole image. This paper proposes SegHSI, an effective and efficient end-to-end HSI segmentation model, alongside a novel training strategy. SegHSI adopts a head-free structure with cluster attention modules and spatial-aware feedforward networks (SA-FFN) for multiscale spatial encoding. Cluster attention encodes pixels through constructed clusters within the HSI, while SA-FFN integrates depth-wise convolution to enhance spatial context. Our training strategy utilizes a student-teacher model framework that combines labeled pixel class information with consistency learning on unlabeled pixels. Experiments on three public HSI datasets demonstrate that SegHSI not only surpasses other state-of-the-art models in segmentation accuracy but also achieves inference time at the scale of seconds, even reaching sub-second speeds for full-image classification. Code is available at https://github.com/huanliu233/SegHSI.

SARSNet—A Novel CNN Approach for SARWater Body Segmentation

This paper presents the SARSNet architecture, developed to address the growing challenges in Synthetic Aperture Radar (SAR) deep learning-based automatic water body extraction. Such a task is riddled with significant challenges, encompassing issues like cloud interference, scarcity of annotated dataset, and the intricacies associated with varied topography. Recent strides in Convolutional Neural Networks (CNNs) and multispectral segmentation techniques offer a promising avenue to address these predicaments. In our research, we propose a series of solutions to elevate the process of water body segmentation. Our proposed solutions span several domains, including image resolution enhancement, refined extraction techniques tailored for narrow water bodies, self-balancing of the class pixel level, and minority class-influenced loss function, all aimed at amplifying prediction precision and streamlining computational complexity inherent in deep neural networks. The framework of our approach includes the introduction of a multichannel Data-Fusion Register, the incorporation of a CNN-based Patch Adaptive Network augmentation method, and the integration of class pixel level balancing and the Tversky loss function. We evaluated the performance of the model using the Sentinel-1 SAR electromagnetic signal dataset from the Earth flood water body extraction competition organized by the artificial intelligence department of Microsoft. In our analysis, our suggested SARSNet was compared to well-known semantic segmentation models, and a comprehensive assessment demonstrates that SARSNet consistently outperforms these models in all data subsets, including training, validation, and testing sets.

SARSNet—A Novel CNN Approach for SARWater Body Segmentation

This paper presents the SARSNet architecture, developed to address the growing challenges in Synthetic Aperture Radar (SAR) deep learning-based automatic water body extraction. Such a task is riddled with significant challenges, encompassing issues like cloud interference, scarcity of annotated dataset, and the intricacies associated with varied topography. Recent strides in Convolutional Neural Networks (CNNs) and multispectral segmentation techniques offer a promising avenue to address these predicaments. In our research, we propose a series of solutions to elevate the process of water body segmentation. Our proposed solutions span several domains, including image resolution enhancement, refined extraction techniques tailored for narrow water bodies, self-balancing of the class pixel level, and minority class-influenced loss function, all aimed at amplifying prediction precision and streamlining computational complexity inherent in deep neural networks. The framework of our approach includes the introduction of a multichannel Data-Fusion Register, the incorporation of a CNN-based Patch Adaptive Network augmentation method, and the integration of class pixel level balancing and the Tversky loss function. We evaluated the performance of the model using the Sentinel-1 SAR electromagnetic signal dataset from the Earth flood water body extraction competition organized by the artificial intelligence department of Microsoft. In our analysis, our suggested SARSNet was compared to well-known semantic segmentation models, and a comprehensive assessment demonstrates that SARSNet consistently outperforms these models in all data subsets, including training, validation, and testing sets.