Semantic Segmentation Model Research Articles

An automatic shoreline extraction method from SAR imagery using DeepLab-v3+ and its versatility

ABSTRACT This study presented an automatic shoreline extraction method from synthetic aperture radar (SAR) imagery using a semantic image segmentation model, DeepLab-v3+. Shorelines extracted from optical satellite images were employed as ground truth. More than 130,000 pairs of labeled images were generated, which substantially enhances the dataset’s objectivity for model training. By using different combination of SAR images from beaches with varying characteristics, several models were constructed. These models were then applied to 15 beaches of Japan to validate their accuracy and versatility. The developed machine learning model demonstrates high accuracy in shoreline extraction, particularly when trained with a diverse set of images from various beaches. This versatility is crucial for the model’s applicability across different coastal environments, underscoring the importance of incorporating a wide range of beach properties into the training dataset.

RGBD-based method for segmenting apparent pores within bridge towers

Abstract As a crucial technology in computer vision, image semantic segmentation is extensively applied to tasks such as detecting apparent defects in concrete, identifying structural cracks, and interpreting facility scenes within infrastructure settings. Challenges such as uneven lighting inside bridge towers and the similarity in color and texture between concrete pore structures and their surrounding areas frequently result in lower segmentation accuracy. This article introduces a multimodal semantic segmentation model incorporating depth information to tackle these challenges. By integrating depth and RGB images as inputs, the model constructs an interactive space and utilizes a cross-attention mechanism along with global context for guided learning, thus enabling precise feature extraction and segmentation. The experimental results demonstrate that this segmentation network excels on a custom-made concrete pore dataset, with a precision rate of 90.88%, recall rate of 87.48%, intersection over union (IoU) of 80.42%, and F1 and Dice coefficient of 89.10% and 89.15%, respectively. These outcomes affirm the network’s effectiveness in segmenting apparent defects in concrete and offer technical support and solutions for image semantic segmentation tasks in settings like bridge towers.

Research on insulator image segmentation and defect recognition technology based on U‐Net and YOLOv7

AbstractThis study focuses on aerial images in power line inspection, using a small sample size and concentrating on accurately segmenting insulators in images and identifying potential “self‐explode” defects through deep learning methods. The research process consists of four key steps: image segmentation of insulators, identification of small connected regions, data augmentation of original samples, and detection of insulator defects using the YOLO v7 model. In this paper, due to the small sample size, sample expansion is considered first. A sliding window approach is adopted to crop images, increasing the number of training samples. Subsequently, the U‐Net neural network model for semantic segmentation is used to train insulator images, thereby generating preliminary mask images of insulators. Then, through connected region area filtering techniques, smaller connected regions are removed to eliminate small speckles in the predicted mask images, obtaining more accurate insulator mask images. The evaluation metric for image recognition, the dice coefficient, is 93.67%. To target the identification of insulator defects, 35 images with insulator defects from the original samples are augmented. These images are input into the YOLO v7 network for further training, ultimately achieving effective detection of insulator “self‐explode” defects.

Nano-particles size measurement based on semantic segmentation via convolution neural network

Automatic recognition and size measurement of nanoparticles in SEM images have been an evolving field of study within the material informatics domain. These methods have been typically assessed using traditional image processing techniques. This paper presents a robust and fully automated pipeline for all researchers, particularly those in chemical engineering, based on image processing techniques and convolution neural networks. We evaluated our method on a dataset of 1,701 2D particle images, employing a three-fold split for training, validation, and testing. The proposed approach integrates semantic segmentation models (U-Net, LinkNet) with size measurement algorithms, achieving high accuracy even in the presence of noise or blurriness. Our results demonstrate the method effectiveness in precisely estimating particle sizes, with a Pearson correlation coefficient of 0.9633 and mean absolute error of 1.287e-06. The method’s robustness against poor-quality images and its fully automated procedures make it a valuable tool for researches in SEM images field.

Containment Control-Guided Boundary Information for Semantic Segmentation

Real-time semantic segmentation is a challenging task in computer vision, especially in complex scenes. In this study, a novel three-branch semantic segmentation model is designed, aiming to effectively use boundary information to improve the accuracy of semantic segmentation. The proposed model introduces the concept of containment control in a pioneering way, which treats image interior elements as well as image boundary elements as followers and leaders in containment control, respectively. Based on this, we utilize two learnable feature fusion matrices in the high-level semantic information stage of the model to quantify the fusion process of internal and boundary features. Further, we design a dedicated loss function to update the parameters of the feature fusion matrices based on the criterion of containment control, which enables fine-grained communication between target features. In addition, our model incorporates a Feature Enhancement Unit (FEU) to tackle the challenge of maximizing the utility of multi-scale features essential for semantic segmentation tasks through the meticulous reconstruction of these features. The proposed model proves effective on the publicly available Cityscapes and CamVid datasets, achieving a trade-off between effectiveness and speed.

Segmentation Network for Multi-Shape Tea Bud Leaves Based on Attention and Path Feature Aggregation

Accurately detecting tea bud leaves is crucial for the automation of tea picking robots. However, challenges arise due to tea stem occlusion and overlapping of buds and leaves, presenting varied shapes of one bud–one leaf targets in the field of view, making precise segmentation of tea bud leaves challenging. To improve the segmentation accuracy of one bud–one leaf targets with different shapes and fine granularity, this study proposes a novel semantic segmentation model for tea bud leaves. The method designs a hierarchical Transformer block based on a self-attention mechanism in the encoding network, which is beneficial for capturing long-range dependencies between features and enhancing the representation of common features. Then, a multi-path feature aggregation module is designed to effectively merge the feature outputs of encoder blocks with decoder outputs, thereby alleviating the loss of fine-grained features caused by downsampling. Furthermore, a refined polarized attention mechanism is employed after the aggregation module to perform polarized filtering on features in channel and spatial dimensions, enhancing the output of fine-grained features. The experimental results demonstrate that the proposed Unet-Enhanced model achieves segmentation performance well on one bud–one leaf targets with different shapes, with a mean intersection over union (mIoU) of 91.18% and a mean pixel accuracy (mPA) of 95.10%. The semantic segmentation network can accurately segment tea bud leaves, providing a decision-making basis for the spatial positioning of tea picking robots.

Insights into the Effects of Tile Size and Tile Overlap Levels on Semantic Segmentation Models Trained for Road Surface Area Extraction from Aerial Orthophotography

Studies addressing the supervised extraction of geospatial elements from aerial imagery with semantic segmentation operations (including road surface areas) commonly feature tile sizes varying from 256 × 256 pixels to 1024 × 1024 pixels with no overlap. Relevant geo-computing works in the field often comment on prediction errors that could be attributed to the effect of tile size (number of pixels or the amount of information in the processed image) or to the overlap levels between adjacent image tiles (caused by the absence of continuity information near the borders). This study provides further insights into the impact of tile overlaps and tile sizes on the performance of deep learning (DL) models trained for road extraction. In this work, three semantic segmentation architectures were trained on data from the SROADEX dataset (orthoimages and their binary road masks) that contains approximately 700 million pixels of the positive “Road” class for the road surface area extraction task. First, a statistical analysis is conducted on the performance metrics achieved on unseen testing data featuring around 18 million pixels of the positive class. The goal of this analysis was to study the difference in mean performance and the main and interaction effects of the fixed factors on the dependent variables. The statistical tests proved that the impact on performance was significant for the main effects and for the two-way interaction between tile size and tile overlap and between tile size and DL architecture, at a level of significance of 0.05. We provide further insights and trends in the predictions of the extensive qualitative analysis carried out with the predictions of the best models at each tile size. The results indicate that training the DL models on larger tile sizes with a small percentage of overlap delivers better road representations and that testing different combinations of model and tile sizes can help achieve a better extraction performance.

Semi-Supervised Building Extraction with Optical Flow Correction Based on Satellite Video Data in a Tsunami-Induced Disaster Scene.

Data and reports indicate an increasing frequency and intensity of natural disasters worldwide. Buildings play a crucial role in disaster responses and damage assessments, aiding in planning rescue efforts and evaluating losses. Despite advances in applying deep learning to building extraction, challenges remain in handling complex natural disaster scenes and reducing reliance on labeled datasets. Recent advances in satellite video are opening a new avenue for efficient and accurate building extraction research. By thoroughly mining the characteristics of disaster video data, this work provides a new semantic segmentation model for accurate and efficient building extraction based on a limited number of training data, which consists of two parts: the prediction module and the automatic correction module. The prediction module, based on a base encoder-decoder structure, initially extracts buildings using a limited amount of training data that are obtained instantly. Then, the automatic correction module takes the output of the prediction module as input, constructs a criterion for identifying pixels with erroneous semantic information, and uses optical flow values to extract the accurate corresponding semantic information on the corrected frame. The experimental results demonstrate that the proposed method outperforms other methods in accuracy and computational complexity in complicated natural disaster scenes.

Artificial intelligence for surgical safety during laparoscopic gastrectomy for gastric cancer: Indication of anatomical landmarks related to postoperative pancreatic fistula using deep learning.

Postoperative pancreatic fistula (POPF) is a critical complication of laparoscopic gastrectomy (LG). However, there are no widely recognized anatomical landmarks to prevent POPF during LG. This study aimed to identify anatomical landmarks related to POPF occurrence during LG for gastric cancer and to develop an artificial intelligence (AI) navigation system for indicating these landmarks. Dimpling lines (DLs)-depressions formed between the pancreas and surrounding organs-were defined as anatomical landmarks related to POPF. The DLs for the mesogastrium, intestine, and transverse mesocolon were named DMP, DIP, and DTP, respectively. We included 50 LG cases to develop the AI system (45/50 were used for training and 5/50 for adjusting the hyperparameters of the employed system). Regarding the validation of the AI system, DLs were assessed by an external evaluation committee using a Likert scale, and the pancreas was assessed using the Dice coefficient, with 10 prospectively registered cases. Six expert surgeons confirmed the efficacy of DLs as anatomical landmarks related to POPF in LG. An AI system was developed using a semantic segmentation model that indicated DLs in real-time when this system was synchronized during surgery. Additionally, the distribution of scores for DMP was significantly higher than that of the other DLs (p < 0.001), indicating the relatively high accuracy of this landmark. In addition, the Dice coefficient of the pancreas was 0.70. The DLs may be used as anatomical landmarks related to POPF occurrence. The developed AI navigation system can help visualize the DLs in real-time during LG.

Development and Testing of Artificial Intelligence-Based Mobile Application to Achieve Cataract Backlog-Free Status in Uttar Pradesh, India

BackgroundUttar Pradesh (UP), the most populous state in India, has about 36 million people aged 50 years or older, spread across more than 100,000 villages. Among them, an estimated 3.5 million suffer from visual impairments, including blindness due to untreated cataracts. To achieve cataract backlog-free status, UP is required to screen this population at the community level and provide treatment to those suffering from cataracts. We envisioned an AI-powered primary screening app utilizing eye images, deployable to frontline health workers for community-level screening. This paper outlines insights gained from developing the AI mobile app “Roshni” for cataract screening. MethodThe AI-based cataract classification model was developed using 13,633 eye images and finalized after three stages of experiments, detecting cataracts in images focused on the eye, iris, and pupil. Overall, 155 experiments were conducted using multiple deep learning algorithms, including ResNet50, ResNet101, YOLOv5, EfficientNetV2, and InceptionV3. We established a minimum threshold of 90 % specificity and sensitivity to ensure the algorithm’s suitability for field use. ResultsThe cataract detection model for eye-focused images achieved 51.9 % sensitivity and 87.6 % specificity, while the model for iris-focused images, using a good/bad iris filter, achieved 52.4 % sensitivity and 93.3 % specificity. The classification model for segmented-pupil images, employing a good/bad pupil filter with UNet-based semantic segmentation model and EfficientNetV2, yielded 96 % sensitivity and 97 % specificity. Field testing with 302 beneficiaries (604 images) showed an overall sensitivity of 86.6 %, specificity of 93.3 %, positive predictive value of 58.4 %, and negative predictive value of 98.5 %. ConclusionThis paper details the development of an AI mobile app designed to facilitate community screening for cataracts by frontline health workers.

High resolution remote sensing recognition of elm sparse forest via deep-learning-based semantic segmentation

Elm (Ulmus pumila L.) sparse forest plays an vital role in maintaining local ecological stability and security in the Otingdag Sandy Land area. Prior studies on elm canopy extraction have predominantly relied on manual parameter configuration, resulting in unsatisfactory levels of generalization. To meet the needs of high-precision and rapid recognition of elm sparse forests in large areas, this study proposed a recognition method for elm sparse forest that orients to high spatial resolution remote sensing imageries, using deep-learning-based semantic segmentation techniques. It can automatically learn features that are conducive to segmenting the canopy of elm trees, and retains good generalization ability on the Gaofen-2 imageries obtained in different regions. First, we constructed a dataset specialized for elm canopy semantic segmentation task, and annotated over 130,000 elm canopies based on Gaofen-2 imageries. In addition, we trained 7 deep-learning semantic segmentation model candidates. Among them, MANet showed the best performance, with its F1-score reaching 81.44%. Lastly, we applied edge detection to the elm canopy coverage area, and automatically extract the elm canopy. The proposed method can provide technical support for the investigation and monitoring of elm sparse forests, while facilitates local desertification prevention efforts in the entire Otingdag Sandy Region.

Research on terahertz image analysis of thin-shell seeds based on semantic segmentation

Assessing crop seed phenotypic traits is essential for breeding innovations and germplasm enhancement. However, the tough outer layers of thin-shelled seeds present significant challenges for traditional methods aimed at the rapid assessment of their internal structures and quality attributes. This study explores the potential of combining terahertz (THz) time-domain spectroscopy and imaging with semantic segmentation models for the rapid and non-destructive examination of these traits. A total of 120 watermelon seed samples from three distinct varieties, were curated in this study, facilitating a comprehensive analysis of both their outer layers and inner kernels. Utilizing a transmission imaging modality, THz spectral images were acquired and subsequently reconstructed employing a correlation coefficient method. Deep learning-based SegNet and DeepLab V3+ models were employed for automatic tissue segmentation. Our research revealed that DeepLab V3+ significantly surpassed SegNet in both speed and accuracy. Specifically, DeepLab V3+ achieved a pixel accuracy of 96.69 % and an intersection over the union of 91.3 % for the outer layer, with the inner kernel results closely following. These results underscore the proficiency of DeepLab V3+ in distinguishing between the seed coat and kernel, thereby furnishing precise phenotypic trait analyses for seeds with thin shells. Moreover, this study accentuates the instrumental role of deep learning technologies in advancing agricultural research and practices.

MixImages: An Urban Perception AI Method Based on Polarization Multimodalities.

Intelligent urban perception is one of the hot topics. Most previous urban perception models based on semantic segmentation mainly used RGB images as unimodal inputs. However, in natural urban scenes, the interplay of light and shadow often leads to confused RGB features, which diminish the model's perception ability. Multimodal polarization data encompass information dimensions beyond RGB, which can enhance the representation of shadow regions, serving as additional data for assistance. Additionally, in recent years, transformers have achieved outstanding performance in visual tasks, and their large, effective receptive field can provide more discriminative cues for shadow regions. For these reasons, this study proposes a novel semantic segmentation model called MixImages, which can combine polarization data for pixel-level perception. We conducted comprehensive experiments on a polarization dataset of urban scenes. The results showed that the proposed MixImages can achieve an accuracy advantage of 3.43% over the control group model using only RGB images in the unimodal benchmark while gaining a performance improvement of 4.29% in the multimodal benchmark. Additionally, to provide a reference for specific downstream tasks, we also tested the impact of different combinations of polarization types on the overall segmentation accuracy. The proposed MixImages can be a new option for conducting urban scene perception tasks.

Utilizing GaoFen-2 derived urban green space information to predict local surface temperature

Urban green spaces (UGS) significantly influence the distribution of surface heat and play a crucial role in regulating surface temperature. However, the quantitative relationship between UGS and surface temperature remains unclear, necessitating further research. This study aims to predict surface temperature based on green space information from GaoFen-2 satellite data. To achieve this, GaoFen-2 data were utilized to obtain spatial distribution and vegetation growth status in Urumqi, Xinjiang. Three machine learning models such as Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting Regression Tree (GBRT) were constructed to predict surface temperature. Results indicated that UGS information extracted from GaoFen-2 data using the U-Net semantic segmentation model successfully predicted surface temperature. Among the three machine learning models, GBRT exhibited the highest predictive accuracy with an Radj2 of 0.81, RMSE of 0.44, and RPD of 2.29, followed by RF (Radj2 of 0.80, RMSE of 0.45, and RPD of 2.22), and SVM (Radj2of 0.79, RMSE of 0.47, and RPD of 2.15), In addition, a variable importance assessment reduced the original 44 variables to 28, maintaining predictive accuracy with the GBRT model achieving an Radj2 of 0.81, RMSE of 0.43, and RPD of 2.3. Our study demonstrates the effectiveness of using vegetation information derived from GaoFen-2 to predict surface temperature. This approach provides valuable recommendations for the layout of UGS in urban areas and serves as a comprehensive reference for urban planning and real estate development.

Towards a Flexible Semantic Guided Model for Single Image Enhancement and Restoration.

Low-light image enhancement (LLIE) investigates how to improve the brightness of an image captured in illumination-insufficient environments. The majority of existing methods enhance low-light images in a global and uniform manner, without taking into account the semantic information of different regions. Consequently, a network may easily deviate from the original color of local regions. To address this issue, we propose a semantic-aware knowledge-guided framework (SKF) that can assist a low-light enhancement model in learning rich and diverse priors encapsulated in a semantic segmentation model. We concentrate on incorporating semantic knowledge from three key aspects: a semantic-aware embedding module that adaptively integrates semantic priors in feature representation space, a semantic-guided color histogram loss that preserves color consistency of various instances, and a semantic-guided adversarial loss that produces more natural textures by semantic priors. Our SKF is appealing in acting as a general framework in the LLIE task. We further present a refined framework SKF++ with two new techniques: (a) Extra convolutional branch for intra-class illumination and color recovery through extracting local information and (b) Equalization-based histogram transformation for contrast enhancement and high dynamic range adjustment. Extensive experiments on various benchmarks of LLIE task and other image processing tasks show that models equipped with the SKF/SKF++ significantly outperform the baselines and our SKF/SKF++ generalizes to different models and scenes well. Besides, the potential benefits of our method in face detection and semantic segmentation in low-light conditions are discussed. The code and pre-trained models have been publicly available at https://github.com/langmanbusi/Semantic-Aware-Low-Light-Image-Enhancement.

The street space planning and design of artificial intelligence-assisted deep learning neural network in the Internet of Things

This study begins by discussing Internet of Things (IoT) technology and analyzing the classification of street space into four types, along with the Green Looking Ratio (GLR). Following this, the Fully Convolutional Network (FCN)-8s framework is employed to construct a street view image semantic segmentation model based on FCN principles. Subsequently, IoT technology is utilized to analyze the proportion of GLR and satisfaction in the street space within the historical urban area of T City. The findings reveal a significant positive correlation (significance level p < 0.05, R2 = 0.919) between the GLR satisfaction score of street view images and the average GLR of the area. Among the four types of street space—life leisure, historical streets, traffic areas, and landscape-style streets—the dissatisfaction rates with GLR are 35 %, 33 %, 20 %, and 18 %, respectively, correlating with varying GLR satisfaction levels. To enhance street space greening, planting ponds and boxes are proposed for "blind spots” and "dead corners,” thereby completing greenery in these areas. These initiatives aim to improve street greening policies, integrate street function zones, and advance the scientific greening of urban streets. The analysis of GLR and satisfaction in street spaces provides valuable insights for refining urban street space greening efforts.

Open-Pit Mining Area Extraction Using Multispectral Remote Sensing Images: A Deep Learning Extraction Method Based on Transformer

In the era of remote sensing big data, the intelligent interpretation of remote sensing images is a key technology for mining the value of remote sensing big data and promoting a number of major applications, mainly including land cover classification and extraction. Among these, the rapid extraction of open-pit mining areas plays a vital role in current practices for refined mineral resources development and management and ecological–environmental protection in China. However, existing methods are not accurate enough for classification, not fine enough for boundary extraction, and poor in terms of multi-scale adaptation. To address these issues, we propose a new semantic segmentation model based on Transformer, which is called Segmentation for Mine—SegMine—and consists of a Vision Transformer-based encoder and a lightweight attention mask decoder. The experimental results show that SegMine enhances the network’s ability to obtain local spatial detail information and improves the problem of disappearing small-scale object features and insufficient information expression. It also better preserves the boundary details of open-pit mining areas. Using the metrics of mIoU, precision, recall, and dice, experimental areas were selected for comparative analysis, and the results show that the new method is significantly better than six other existing major Transformer variants.

Utilizing Silhouette and Head Information for Improved Cloth-changing Person Re-Identification

In recent years, significant progress has been made in person re-identification (ReID). However, as the time span increases, the phenomenon of changing clothes often occur, and the performance of methods based on person re-identification may decrease in this case. In this article, we focus on solving the problem of cloth-changing person re-identification from a single RGB image, which is more challenging, and improving the accuracy and robustness of traditional methods for recognition. We propose a cloth-changing Feature Regularization learning framework by fusing Silhouette and Head information (SHFR), a two-stream framework which transfers the silhouette information and head information learned by auxiliary stream to the main stream and supplements features unrelated to clothing. Specifically, the main stream approach we adopt is the traditional cloth-changing person re-identification network. In the auxiliary stream, we use the DeepLabV3 semantic segmentation model to extract person silhouette features, and the FCHD fully convolutional head detection model to extract head information with advanced semantic information. We concatenate the head information and background information into the silhouette as input for BackBone to continue training the network. In order to utilize silhouette information and head information, we regularize the cloth-changing features on both main stream and auxiliary stream, allowing main stream model to pay more attention to features unrelated to clothing and improve the accuracy of model recognition. We conduct extensive experiments on our model on the PRCC dataset, and the results show that our method has highly competitive performance.

CvT-UNet: A weld pool segmentation method integrating a CNN and a transformer

In the acquisition of weld seam images, the presence of light and impurity interference often leads to weak edge problems in weld seam images, making the identification of weld seam defects challenging. In such scenarios, segmentation processing must be performed specifically at the fusion pool of the weld seam. In recent years, Transformers, which leverage self-attention mechanisms, have demonstrated superior sequence modeling capabilities compared to convolutional neural networks (CNNs). The Transformer excels at weighting different positions within a sequence, enabling effective handling of long-range dependencies. However, Transformers lack the ability of CNNs to extract local information. To address this limitation, this paper introduces a novel CvT-UNet semantic segmentation model that combines the strengths of both the convolutional neural network and the transformer. The model combines the global context information capabilities of the Transformer with the spatial information advantages provided by the convolutional neural network. The proposed model employs an encoder-decoder architecture by building a U-shaped network with designed CvT blocks. Depthwise separable convolutions are utilized to balance parameter quantities, and the skip-connection module is redesigned to enable the model to achieve more precise segmentation of weld seam fusion pools with fewer parameters. Experimental evaluations on diverse weld seam datasets under different environmental conditions demonstrate promising results. The model achieves average intersection over union (IoU) values of 93.75 %, 88.31 %, and 90.86 %, respectively. Specifically, on an automotive weld seam dataset, CvT-UNet outperforms the purely convolutional neural network structure UNet3+, with a 1.63 % increase in the mean IoU (MIOU) and a 1.39 % increase in the mean pixel accuracy (MPA). Compared to the TransUNet fusion model, CvT-UNet results in a 0.33 % increase in the MIOU and a 0.6 % increase in the MPA. Furthermore, compared to the state-of-the-art curve segmentation model LIOT, CvT-UNet achieves a 1.94 % increase in the MIOU and a 0.61 % increase in the MPA, demonstrating favorable segmentation performance.

Winter Wheat Mapping Method Based on Pseudo-Labels and U-Net Model for Training Sample Shortage

In recent years, the semantic segmentation model has been widely applied in fields such as the extraction of crops due to its advantages such as strong discrimination ability, high accuracy, etc. Currently, there is no standard set of ground true label data for major crops in China, and the visual interpretation process is usually time-consuming and laborious. The sample size also makes it difficult to support the model to learn enough ground features, resulting in poor generalisation ability of the model, which in turn makes the model difficult to apply in fine extraction tasks of large-area crops. In this study, a method to establish a pseudo-label sample set based on the random forest algorithm to train a semantic segmentation model (U-Net) was proposed to perform winter wheat extraction. With the help of the GEE platform, Winter Wheat Canopy Index (WCI) indicators were employed in this method to initially extract winter wheat, and training samples (i.e., pseudo labels) were built for the semantic segmentation model through the iterative process of “generating random sample points—random forest model training—winter wheat extraction”; on this basis, the U-net model was trained with multi-time series remote sensing images; finally, the U-Net model was employed to obtain the spatial distribution map of winter wheat in Henan Province in 2022. The results illustrated that: (1) Pseudo-label data were constructed using the random forest model in typical regions, achieving an overall accuracy of 97.53% under validation with manual samples, proving that its accuracy meets the requirements for U-Net model training. (2) Utilizing the U-Net model, U-Net++ model, and random forest model constructed based on pseudo-label data for 2022, winter wheat mapping was conducted in Henan Province. The extraction accuracy of the three models is in the order of U-Net model &gt; U-Net++ model &gt; random forest model. (3) Using the U-Net model to predict the winter wheat planting areas in Henan Province in 2019, although the extraction accuracy decreased compared to 2022, it still exceeded that of the random forest model. Additionally, the U-Net++ model did not achieve higher classification accuracy. (4) Experimental results demonstrate that deep learning models constructed based on pseudo-labels exhibit higher classification accuracy. Compared to traditional machine learning models like random forest, they have higher spatiotemporal adaptability and robustness, further validating the scientific and practical feasibility of pseudo-labels and their generation strategies, which are expected to provide a feasible technical pathway for intelligent extraction of winter wheat spatial distribution information in the future.