Semantic segmentation of target objects in power transmission line corridor point cloud scenes is a crucial step in powerline tree barrier detection. The massive quantity, disordered distribution, and non-uniformity of point clouds in power transmission line corridor scenes pose significant challenges for feature extraction. Previous studies have often overlooked the core utilization of spatial information, limiting the network’s ability to understand complex geometric shapes. To overcome this limitation, this paper focuses on enhancing the deep expression of spatial geometric information in segmentation networks and proposes a method called BDF-Net to improve RandLA-Net. For each input 3D point cloud data, BDF-Net first encodes the relative coordinates and relative distance information into spatial geometric feature representations through the Spatial Information Encoding block to capture the local spatial structure of the point cloud data. Subsequently, the Bilinear Pooling block effectively combines the feature information of the point cloud with the spatial geometric representation by leveraging its bilinear interaction capability thus learning more discriminative local feature descriptors. The Global Feature Extraction block captures the global structure information in the point cloud data by using the ratio between the point position and the relative position, so as to enhance the semantic understanding ability of the network. In order to verify the performance of BDF-Net, this paper constructs a dataset, PPCD, for the point cloud scenario of transmission line corridors and conducts detailed experiments on it. The experimental results show that BDF-Net achieves significant performance improvements in various evaluation metrics, specifically achieving an OA of 97.16%, a mIoU of 77.48%, and a mAcc of 87.6%, which are 3.03%, 16.23%, and 18.44% higher than RandLA-Net, respectively. Moreover, comparisons with other state-of-the-art methods also verify the superiority of BDF-Net in point cloud semantic segmentation tasks.

In recent years, accurate field monitoring has been a research hotspot in the domains of aerial remote sensing and satellite remote sensing. In view of this, this study proposes an innovative cross-platform super-resolution reconstruction method for remote sensing images for the first time, aiming to make medium-resolution satellites capable of field-level detection through a super-resolution reconstruction technique. The progressive growing generative adversarial network (PGGAN) model, which has excellent high-resolution generation and style transfer capabilities, is combined with a deep residual network, forming the Res-PGGAN model for cross-platform super-resolution reconstruction. The Res-PGGAN architecture is similar to that of the PGGAN, but includes a deep residual module. The proposed Res-PGGAN model has two main benefits. First, the residual module facilitates the training of deep networks, as well as the extraction of deep features. Second, the PGGAN structure performs well in cross-platform sensor style transfer, allowing for cross-platform high-magnification super-resolution tasks to be performed well. A large pre-training dataset and real data are used to train the Res-PGGAN to improve the resolution of Sentinel-2’s 10 m resolution satellite images to 0.625 m. Three evaluation metrics, including the structural similarity index metric (SSIM), the peak signal-to-noise ratio (PSNR), and the universal quality index (UQI), are used to evaluate the high-magnification images obtained by the proposed method. The images generated by the proposed method are also compared with those obtained by the traditional bicubic method and two deep learning super-resolution reconstruction methods: the enhanced super-resolution generative adversarial network (ESRGAN) and the PGGAN. The results indicate that the proposed method outperforms all the comparison methods and demonstrates an acceptable performance regarding all three metrics (SSIM/PSNR/UQI: 0.9726/44.7971/0.0417), proving the feasibility of cross-platform super-resolution image recovery.

Crop phenology is one of the most critical physiological attributes of agricultural crops, serving as a direct indicator of growth status throughout the developmental cycle. With the advancement of phenological research, satellite remote sensing has emerged as a primary monitoring tool due to its large spatial coverage and convenient data acquisition. However, high-resolution remote sensing satellites, which are essential for precise phenological observations, often have long revisit intervals. Additionally, adverse atmospheric conditions such as cloud cover frequently compromise the usability of images on multiple dates. As a result, high-resolution time-series data for crop phenology monitoring are typically sparse, limiting the ability to capture rapid phenological changes during the growing season.To address this challenge, this study focuses on paddy and dryland fields as experimental sites and proposes a novel method for filling temporal gaps in remote sensing data using generative image processing techniques. Specifically, a lightweight super-resolution Generative Adversarial Network (GAN) is developed for image reconstruction. Using the reconstructed dataset, dense time-series monitoring and phenological metric extraction were conducted throughout the crop growing season.(1) The proposed super-resolution reconstruction method achieves structural similarity index (SSIM) and peak signal-to-noise ratio (PSNR) values of 0.834 and 28.69, respectively, outperforming mainstream approaches in reconstructing heterogeneous remote sensing data.(2) Following temporal reconstruction, the revisit intervals of remote sensing imagery for the two test sites improved from 6.40 and 6.63 days to 5.70 and 5.88 days, respectively. To further analyze phenological metrics, four smoothing techniques were applied, among which Savitzky–Golay filtering yielded the most accurate and robust results. Although discrepancies were observed between the results obtained using the reconstructed data and those based on the original datasets, the proposed method demonstrated smaller deviations from benchmark datasets. Compared with conventional interpolation-based gap-filling approaches, the framework demonstrated marked improvements in the accuracy of phenological extraction, while also delivering superior spatial resolution and robustness relative to the Harmonized Landsat and Sentinel (HLS) dataset. Experimental results confirm that the proposed approach effectively fills temporal gaps in satellite imagery, enhances data continuity, accurately captures key phenological turning points, and enables precise crop phenology monitoring at high spatial and temporal resolution.

High-resolution Earth observation data is indispensable for remote sensing applications. However, achieving imagery with both high spatial (10m) and temporal (every 2 days) resolution presents significant challenges due to satellite, technological, and budgetary constraints. This paper introduces a novel approach by integrating MODIS and Sentinel data through an enhanced deep learning framework. The existing Enhanced Deep Convolutional Spatiotemporal Fusion Network (EDCSTFN) is augmented by incorporating multi-scale convolutions within the encoder to more effectively capture Landsat features. Additionally, standard convolutions are supplanted by a residual encoder to facilitate more comprehensive feature extraction.A Squeeze-and-Excitation (SE) channel attention module is integrated to enhance nonlinear processing capabilities and optimize channel interrelationships. These modifications have led to measurable improvements in the model's performance, including an increase in the structural similarity index by 0.2%-0.3% and a reduction in normalized global error by 0.3%-0.4%. The newly developed Modified Integrated Enhanced Deep Convolutional Spatiotemporal Fusion Network (MIEDCSTFN) utilizes the processed Landsat data from EDCSTFN as input for the DSTFN model, generating 10m resolution data.Rigorous comparative analyses and validation against multiple models confirm the enhanced accuracy and robustness of MIEDCSTFN. The resultant imagery closely approximates actual observations, making significant contributions to remote sensing time-series analysis applications such as crop monitoring, environmental change detection, and resource management. This breakthrough not only holds substantial social value but also offers extensive application potential across diverse fields.

Net Primary Productivity (NPP) is a vital indicator for evaluating the carbon source and sink capacities of ecosystems, significantly influencing assessments of agricultural productivity and carbon cycle studies. Accurately estimating NPP in the agricultural sector, however, remains challenging. This research addresses the challenge by refining the estimation of the Fraction of Photosynthetically Active Radiation (FPAR) within the CASA model, introducing a novel methodology that significantly improves the accuracy of NPP estimation and, when applied to remote sensing imagery covering a broad region, demonstrates strong potential for large-scale crop NPP monitoring. We employed high-resolution Sentinel-2 satellite imagery and the Recursive Feature Elimination algorithm to extract FPAR-related features from 15 vegetation indices. The FPAR was subsequently estimated using a Convolutional Neural Network, leading to a dramatic decrease in the Root Mean Square Error (RMSE) from 0.2040 to 0.0020. The prediction errors for the improved model ranged from 0.0001 to 0.0092, with a mean absolute error (MAE) below 0.01. These values reflect the distribution of absolute residuals and indicate a substantial enhancement in accuracy over traditional methods. This improved FPAR estimation method was subsequently integrated into the CASA model. Compared to field-measured NPP data, the optimized model reduced the Mean Absolute Percentage Error (MAPE) from 28.92% to 20.31%. The MAPE values across the test samples ranged between 15% and 25%, indicating a significant improvement in model reliability. The optimized CASA model performs well in estimating net primary productivity (NPP) of crops, providing strong support for agricultural decision-making and future research on large-scale productivity and carbon cycling.

In recent years, super-resolution (SR) reconstruction techniques for remote sensing imagery has attracted extensive attention due to its advantages of a low cost and flexible and convenient application. Unlike conventional SR tasks, the SR reconstruction of remote sensing images under real-world conditions is usually accompanied by a large number of objective unfavorable factors, including high levels of noise, cloud occlusion, and variability across different sensors. Consequently, real-scene SR reconstruction constitutes a multitask problem, yet existing SR approaches seldom address these challenges in a unified framework. To overcome these limitations, this study proposes a multitask SR reconstruction network named multitask SR reconstruction generative adversarial network (GAN) specifically designed for real-world remote sensing imagery. The proposed network adopts the GAN model’s structure, and integrates three dedicated modules within the generator: a deep feature extraction module, a spectral transfer module, and a thin cloud removal module. Within this framework, we design a novel Swin Transformer module to enhance spatial diversity and training stability in image reconstruction. Moreover, we introduce a framework termed the efficient feature aggregation architecture, which ensures effective feature fusion while enabling seamless collaboration among these components. Together, these components enable effective adaptation to cross-sensor SR reconstruction tasks under complex real-world conditions. The proposed method is verified by both quantitative and qualitative analyses. The SR reconstruction results obtained on multiple datasets show that the proposed method is superior to the existing mainstream methods.