Weaklier-Supervised Semantic Segmentation with Pyramid Scene Parsing Network

Polarimetric synthetic aperture radar (PolSAR) image semantic segmentation is currently of great importance for synthetic aperture radar (SAR) image interpretation, especially in agricultural applications. Several convolutional neural networks (CNNs) have been implemented for SAR image semantic segmentation in urban or land cover applications. However, existing CNNs often break one semantic area into several pieces or confuse the adjacent semantic area in a PolSAR image. To address these issues uniformly, a refined pyramid scene parsing network (PSPNet) is proposed for PolSAR image semantic segmentation in an agricultural areas. Compared to conventional PSPNet architecture, the refined PSPNet adopts a multilevel feature fusion design in its decoder to effectively exploit the features learned from its different encoder branches. Besides, a polarimetric channel attention module is incorporated into the network to capture rich polarimetric features in a PolSAR image. Furthermore, an edge-aware loss function is devised to guide the network to refine pixel-level edge information directly from semantic segmentation prediction, separating easily confused agriculture area with sharp contours. Experimental results on one airborne millimeter-wave PolSAR dataset verify that the proposed network achieves promising semantic segmentation accuracy and preferable spatial consistency.

An image semantic segmentation algorithm using fully convolutional network (FCN) integrated with the recently proposed simple linear iterative clustering (SLIC) that is based on boundary term (BSLIC) is developed. To improve the segmentation accuracy, the developed algorithm combines the FCN semantic segmentation results with the superpixel information acquired by BSLIC. During the combination process, the superpixel semantic annotation is newly introduced and realized by the four criteria. The four criteria are used to annotate a superpixel region, according to FCN semantic segmentation result. The developed algorithm can not only accurately identify the semantic information of the target in the image, but also achieve a high accuracy in the positioning of small edges. The effectiveness of our algorithm is evaluated on the dataset PASCAL VOC 2012. Experimental results show that the developed algorithm improved the target segmentation accuracy in comparison with the traditional FCN model. With the BSLIC superpixel information that is involved, the proposed algorithm can get 3.86%, 1.41%, and 1.28% improvement in pixel accuracy (PA) over FCN-32s, FCN-16s, and FCN-8s, respectively.

Semantic segmentation is a crucial approach for remote sensing interpretation. High-precision semantic segmentation results are obtained at the cost of manually collecting massive pixelwise annotations. Remote sensing imagery contains complex and variable ground objects and obtaining abundant manual annotations is expensive and arduous. The semi-supervised learning (SSL) strategy can enhance the generalization capability of a model with a small number of labeled samples. In this study, a novel semi-supervised adversarial semantic segmentation network is developed for remote sensing information extraction. A multiscale input convolution module (MICM) is designed to extract sufficient local features, while a Transformer module (TM) is applied for long-range dependency modeling. These modules are integrated to construct a segmentation network with a double-branch encoder. Additionally, a double-branch discriminator network with different convolution kernel sizes is proposed. The segmentation network and discriminator network are jointly trained under the semi-supervised adversarial learning (SSAL) framework to improve its segmentation accuracy in cases with small amounts of labeled data. Taking building extraction as a case study, experiments on three datasets with different resolutions are conducted to validate the proposed network. Semi-supervised semantic segmentation models, in which DeepLabv2, the pyramid scene parsing network (PSPNet), UNet and TransUNet are taken as backbone networks, are utilized for performance comparisons. The results suggest that the approach effectively improves the accuracy of semantic segmentation. The F1 and mean intersection over union (mIoU) accuracy measures are improved by 0.82–11.83% and 0.74–7.5%, respectively, over those of other methods.

Superpixel algorithm aims to semantically group neighboring pixels into a coherent region. It could significantly boost the performance of the subsequent vision processing task such as image segmentation. Recently, the work simple linear iterative clustering (SLIC) [1] has drawn huge attention for its state-of-the-art segmentation performance and high computational efficiency. However, the performance of SLIC is dramatically degraded for noisy images. In this work, we propose three measures to improve the robustness of SLIC against noise: 1) a new pixel intensity distance measurement is designed by explicitly considering the within-cluster noise variance; 2) the spatial distance measurement is refined by exploiting the variation of pixel locations in a cluster; and 3) a noise-robust estimator is proposed to update the cluster centers by excluding the possible outliers caused by noise. Extensive experimental results on synthetic noisy images validate the effectiveness of those improvements. In addition, we apply the proposed noise-robust SLIC to superpixel-based noise level estimation task to demonstrate its practical usage.

For better analyzing and adopting superpixel methods in image segmentation, an improved Simple Linear Iterative Clustering (SLIC) algorithm is put forward, which sustains that homogeneous regions show high consistence during clustering. Firstly, a primary iteration of SLIC approach is employed to generate uniform superpixels. Then an update strategy is introduced to determine whether to recluster the candidate region. In homogeneous regions, a cluster center often shows relatively small deviation of last position in spatial-color feature space when local iteration occurs, which can be regarded as redundant computation and be reduced. Experimental results show that the strategy brings a remarkable increased efficiency for traditional SLIC without significant affection on segmentation results.

This paper investigates the application of the Simple Linear Iterative Clustering (SLIC) algorithm for complex object image segmentation, on the example of images of human body injuries. The study solves the problem of the lack of quantitative evidence regarding SLIC's algorithm performance in high-precision area and boundary assessment of the lesion on a digital image of a human body with a wound injury on it. A comprehensive methodology is developed to evaluate SLIC's algorithm efficacy across various complex images and image resolutions. The research utilizes a combined dataset of 3696 wound images from the Foot Ulcer Segmentation Challenge (FUSeg) and WoundSeg datasets. Bayesian optimization is utilized to fine-tune SLIC algorithm hyperparameters, focusing on the number of segments and compactness. Results indicate that SLIC algorithm demonstrates consistent performance across different implementations, achieving Dice scores around 0.84 and Soft Boundary F1 scores around 0.55. The study reveals that the optimal number of segments for SLIC algorithm can be defined relative to the spatial dimensions of the input image, with maximal image dimension *2 being the most effective value. A thorough analysis of various segmentation metrics is conducted, including IoU, Dice Score, and Boundary F1 Score. The research introduces and employs the Soft Boundary F1 Score – modification of Boundary F1 Score, a novel metric designed to provide a more nuanced evaluation of boundary detection accuracy while offering a smoother optimization landscape. This metric proves particularly valuable in assessing the performance of SLIC algorithm in image with complex objects on them segmentation tasks. Importantly, this research presents an idealized SLIC-based segmentation approach, where superpixels are optimally combined using ground-truth masks to establish an upper bound of performance. This idealized SLIC algorithm segmentation is compared with a pre-trained on the FUSeg dataset UNet model, showcasing superior generalization capability across diverse wound types. On the WoundSeg dataset, the idealized SLIC algorithm approach achieved a Dice score of 0.84, significantly outperforming the UNet model (0.12 Dice score). As a result, this study provides valuable insights for improving complex objects segmentation methods and highlights the need for further research on developing effective methods for superpixel classification in real-world scenarios. The findings also highlight the potential of SLIC-based approaches in addressing the challenges of diverse data types and limited training data.

Semi-supervised image segmentation in order to get a divided image from the tag portion. In this paper, a new semi-supervised image segmentation framework with global boundaries of probability (gPb) and simple linear iterative clustering (SLIC) is proposed. We use SLIC to perform image segmentation of super pixels, and we have constructed a graph, in which each vertex is a superpixel. Then we utilize gPb to define the edge weights for the graph. Semi-supervised learning based image (GSSL) method, and the use of labeled samples to calculate anchor anchor category, the category may be further determined by the anchor point of the sample. The results of the experiments certify the robustness and effectiveness of the method proposed in this paper.

One promising approach for pixel-wise semantic segmentation is based on higher-order Conditional Random Fields (CKFs). We aim to selectively choose segments for the higher-order CRFs in semantic segmentation. To this end, we formulate the selection as an optimization problem. We propose three optimization criteria in relation to the selected segments, namely: a) averaged goodness, b) coverage area and c) non-overlapped area. Essentially, we desire to have best segments with maximum coverage area and maximum non-overlapped area. We apply two evolutionary optimization algorithms, namely: the genetic algorithm (GA) and the particle swarm optimization (PSO). The goodness of segments is estimated using the Latent Dirichlet Allocation approach. Experiment results show that semantic segmentation with GA-or-PSO-selected segments yields competitive semantic segmentation accuracy in comparison to that of naively using all segments. Moreover, the fewer number of segments used in semantic segmentation speeds up its computation time up to six times faster.

Low contrast images and blurriness pose challenge in the over‐segmentation of image, which increases model complexities. In this work, a novel hybrid dermoscopic skin‐lesion segmentation method, namely SLICACO, is proposed incorporating the simple linear iterative clustering (SLIC) and ant colony optimization (ACO) algorithms. The working of proposed method is multifold. First, over‐segmentation of preprocessed image is generated using SLIC super‐pixel technique. Second, clusters of super‐pixels generated by SLIC are used by ACO with the pixels of similar intensity for edge detection and seek for the optimum pathway in a strained zone. Third, lesion area is segmented using the Convex Hull and Thresholding. Fourth, Erosion Filtering is used to obtain the final segmented image. The performance of SLICACO is assessed on five benchmark dermatoscopic datasets and compared with deep learning models to test its generalizing behavior. Promising results are obtained on the PH2 archive data set with an accuracy of 95.9%.

Weed management continues to be a significant challenge in modern agriculture, primarily due to the aggressive growth patterns of weeds and their direct competition with crops for essential resources such as light, water, and nutrients. Although recent developments in precision agriculture have led to the emergence of automated weed detection systems aimed at reducing operational costs and decreasing reliance on chemical herbicides, achieving accurate crop–weed segmentation remains a persistent difficulty. This is largely attributed to high visual similarity between crops and weeds, coupled with variations in illumination and field conditions. To address these challenges, Convolutional Neural Networks (CNNs) have been increasingly adopted for their capability to perform end-to-end, pixel-level classification, particularly when leveraging multi-spectral imagery. In this context, PSPEdgeWeedNet is proposed, a novel edge-aware deep learning architecture tailored for precise semantic segmentation of crops and weeds within peanut cultivation fields. Distinct from the conventional Pyramid Scene Parsing Network (PSPNet) and its boundary-aware variant developed as a baseline in this research, PSPEdgeWeedNet introduces a dedicated edge detection branch. This branch is specifically engineered to enhance boundary localization and improve delineation between adjacent vegetation classes. In post-processing, Conditional Random Fields (CRFs) are used to slightly enhance the segmentation results around object boundaries. Additionally, all models were trained on a curated peanut field dataset using class-weighted loss functions to effectively address inherent class imbalance. Comprehensive experimental evaluations reveal that PSPEdgeWeedNet significantly outperforms existing state-of-the-art architectures including PSPNet, SegNet, UNet, DeepLabv3, Swin-Unet, and light weight transformer model based on ViT across multiple performance metrics such as Intersection over Union (IoU), precision, recall, and F1-score. These results highlight the critical role of incorporating edge-aware mechanisms within semantic segmentation frameworks, thereby enhancing the robustness and accuracy of automated weed detection systems in complex, real-world agricultural environments.

In recent years, regional algorithms have shown great potential in the field of synthetic aperture radar (SAR) image segmentation. However, SAR images have a variety of landforms and a landform with complex texture is difficult to be divided as a whole. Due to speckle noise, traditional over-segmentation algorithm may cause mixed superpixels with different labels. They are usually located adjacent to two areas or contain more noise. In this paper, a new semantic segmentation method of SAR images based on texture complexity analysis and key superpixels is proposed. Texture complexity analysis is performed and on this basis, mixed superpixels are selected as key superpixels. Specifically, the texture complexity of the input image is calculated by a new method. Then a new superpixels generation method called neighbourhood information simple linear iterative clustering (NISLIC) is used to over-segment the image. For images with high texture complexity, the complex areas are first separated and key superpixels are selected according to certain rules. For images with low texture complexity, key superpixels are directly extracted. Finally, the superpixels are pre-segmented by fuzzy clustering based on the extracted features and the key superpixels are processed at the pixel level to obtain the final result. The effectiveness of this method has been successfully verified on several kinds of images. Comparing with the state-of-the-art algorithms, the proposed algorithm can more effectively distinguish different landforms and suppress the influence of noise, so as to achieve semantic segmentation of SAR images.

Semantic segmentation is a fundamental task in image processing and computer vision domains. Therefore, it is a core component of most emerging industry applications such as medical imaging, autonomous driving, and agriculture. The last five years have witnessed a huge growth in the computer vision domain with the introduction of Generative Adversarial Networks (GANs) due to their ability to learn deep representations by a competitive process between two neural networks called the Generator the Discriminator. Neural Architecture Search (NAS) is undergoing a revolution aimed at automating the architectural engineering process. This paper discusses implementing a novel framework for semantic image segmentation called iSegmentor using NAS and GANs. The proposed approach is the first framework that utilizes NAS in GANs for semantic image segmentation to the best of our knowledge. The iSegmentor framework was evaluated using mean Intersection over Union (mIoU) and Mean Pixel Accuracy (MPA) metrics. The iSegmentor was mainly tested and evaluated for the autonomous driving domain. With the analysis of the past five-year literature, Cityscapes and CamVid datasets were chosen. The iSegmentor achieved 81.73 MPA and 71.90 mIoU for the Cityscapes dataset, whereas CamVid achieved 80.41 MPA and 70.63 mIoU. Further, we present future avenues that we identified with the conducted experiments to assist other researchers to extend the research area of NAS in GANs in semantic segmentation.

Simple Linear Iterative Clustering (SLIC) algorithm is increasingly applied to different kinds of image processing because of its excellent perceptually meaningful characteristics. In order to better meet the needs of medical image processing and provide technical reference for SLIC on the application of medical image segmentation, two indicators of boundary accuracy and superpixel uniformity are introduced with other indicators to systematically analyze the performance of SLIC algorithm, compared with Normalized cuts and Turbopixels algorithm. The extensive experimental results show that SLIC is faster and less sensitive to the image type and the setting superpixel number than other similar algorithms such as Turbopixels and Normalized cuts algorithms. And it also has a great benefit to the boundary recall, the robustness of fuzzy boundary, the setting superpixel size and the segmentation performance on medical image segmentation.

Reversible image watermarking schemes are used to protect ownership and copyrights of digital images. This paper proposes a novel reversible image watermarking scheme based on adaptive image interpolation, segmentation and additive prediction error expansion (PEE). Proposed interpolation comprises of weighted average of neighboring pixels by allocating higher and lower weights to less and more distant neighboring pixel values respectively. The proposed adaptive image interpolation focuses on detection of edges thus minimizing artifacts imposed by interpolation. The idea of embedding varying amount of watermark bits in different image segments has been explored. Simple linear iterative clustering (SLIC) based image segmentation is performed to separate very sharp, sharp, smooth and very smooth regions in image. Higher number of watermark bits are embedded in sharp regions by using additive prediction error expansion embedding technique. Simulations of proposed and existing techniques were performed on different images and compared using embedding capacity (EC), peak signal to noise ratio (PSNR), computational efficiency, image quality, mean square error (MSE), normalized cross correlation (NCC) and structural similarity index (SSIM). The experimental results show that proposed scheme achieves better results in terms of EC, PSNR, computational efficiency, image quality, MSE, NCC and SSIM as compared to existing techniques.

Research on image classification sparked the latest deep-learning boom. Many downstream tasks, including semantic segmentation, benefit from it. The state-of-the-art semantic segmentation models are all based on deep learning, and they sometimes make some semantic mistakes. In a semantic segmentation dataset with a small number of categories, images are often collected from a single scene, and there is a close semantic connection between any two categories. However, in the semantic segmentation dataset collected from multiple scenes, two categories may be irrelevant. The probability of objects in one category appearing next to objects in other categories is different, which is the basis of the paper. Semantic segmentation methods need to solve two problems of positioning and classification. This paper is dedicated to correcting those clearly wrong classifications that are contrary to reality. Specifically, we first calculate the relevancy between different class pairs. Then, based on this knowledge, we infer the category of a connected component according to the relationships of this connected component with its surrounding connected components and correct the obviously wrong classifications made by a deep learning semantic segmentation model. Several well-performing deep learning models are experimented on two challenging public datasets in the field of semantic image segmentation. Our proposed method improves the performance of UPerNet, OCRNet and SETR from 40.7%, 43% and 48.64% to 42.07%, 44.09% and 49.09% mean IoU on the ADE20K validation set, and the performance of PSPNet, DeepLabV3 and OCRNet from 37.26%, 37.3% and 39.5% to 38.93%, 38.95% and 40.63% mean IoU on the COCO-Stuff dataset, which shows the effectiveness of the method.