Shulei Wu

Introduction: Mangrove extent maps lack the taxonomic information needed for carbon assessment, restoration evaluation, and management. Taxon mapping at 10 m remains constrained by spectral similarity, mixed pixels, limited reference coverage, and spatial dependence. Methods: We developed the Inverted-Residual Spectral One-Dimensional Convolutional Neural Network (IR-Spec1D) as a compact classifier for a 14-feature Sentinel-1/2 optical and radar vector. To capture nonlinear optical–radar relationships compactly, inverted-residual and depthwise-separable transformations learn feature interactions, while Efficient Channel Attention emphasizes informative learned channels. Spectral-Domain Dropout reduces dependence on individual inputs through training-time feature masking. The workflow combines this configuration with spatial evaluation, post hoc temperature scaling, predictive-entropy analysis, and frozen transfer assessment. The six-category source taxonomy includes the B. sexangula / B. gymnorhiza complex. Six classifiers were evaluated using April 2024 observations from the selected Dongzhai Harbor study area. Results: IR-Spec1D achieved the highest mean overall accuracy (0.9720) and macro-F1 (0.9406) under five spatially grouped cross-validation folds. Performance remained comparatively high with a 100-m exclusion buffer. Held-out source-domain diagnostics showed that temperature scaling reduced mean expected calibration error from 0.0346 to 0.0125 without changing categorical predictions. Predictive entropy ranked cross-validated source-domain errors with an AUROC of 0.8808. The frozen ensemble retained an overall accuracy of 0.9450 in both same-site temporal evaluations, whereas four-category closed-set geographic transfer to Gaoqiao yielded 0.7118. Discussion: The combined evidence characterizes taxon discrimination, held-out probability behavior, and performance retention under specified transfer conditions within a common mangrove-mapping workflow.

As a typical blue carbon ecosystem, mangroves play a vital role in the global carbon cycle and coastal ecological security. Accurately estimating their aboveground biomass (AGB) is crucial for carbon stock accounting and ecosystem monitoring. However, mangrove AGB estimation faces two core challenges: complex, inaccessible habitats limiting high-quality traditional sampling, and inherent limitations of single-source remote sensing data with inadequate multi-source feature fusion—restricting deep learning applications and relevant research. This study adopted UAV-LiDAR for high-precision sampling and proposed an improved CNN, Residual Attention Pyramid CNN (RAP CNN), with distinct advantages in tackling these challenges: the SE module enhances biomass-sensitive features, suppresses redundancy, and resolves inadequate multi-source fusion; the SPP module captures multi-scale spatial context to improve estimation stability in heterogeneous areas; residual connections mitigate gradient vanishing and strengthen deep feature ex traction. Integrating Sentinel-1/2 multi-source data, it upgrades from simple feature stacking to intelligent synergy of multisource information. Innovatively combining UAV-LiDAR-derived high-precision samples (tree height and DBH via individual tree segmentation) with species classification results and allo metric equations, we constructed species-level biomass samples, overcoming traditional sampling bottlenecks. Results show RAP CNN achieves R2 = 0.7645, RMSE = 8.3903 Mg/ha on the training set and R2 = 0.7576, RMSE = 13.0467 Mg/ha on the validation set, outperforming traditional methods like RF and XGBoost. These findings verify RAP-CNN's effectiveness and UAV-LiDAR's key role, providing valuable technical support for carbon stock assessment and ecological protection.

Aiming at the problems of low scrambling dimension, complex diffusion mechanism and weak anti-exploit competency in multi-image encryption, this paper proposes a multi-image enctyption algorithm based on three-dimensional simplified quadratic sine map (3D-SQSM) and three-dimensional coordinate matrix. First, a new 3D-SQSM chaotic system is designed, and its strong chaotic characteristics and high stochasticity in a wide parameter interval are verified by bifurcation diagram, Lyapunov exponent, 0–1 test and NIST stochastic test. Second, a global scrambling mechanism based on three-dimensional coordinate matrix is proposed: stack multiple images into a three-dimensional matrix, take advantage of the chaotic sequence generated by 3D-SQSM to construct a three-dimensional coordinate point set, and generate a three-dimensional index matrix through global dictionary order sequencing, so as to realize the cross-image and cross-dimensional global reflow of picture element positioning. Subsequently, in the diffusion stage, multi-channel XOR Actions based on chaotic sequences are adopted, combined with the SHA-256 hash function to generate a secret key strongly related to the plaintext, ensuring that the algorithm has strong resistance to chosen-plaintext exploit. Simulation studies and security analysis demonstrate that the proposed method possesses a sufficiently large key space, exhibits high sensitivity to minor key variations, achieves uniform distribution in the histogram of encrypted images, reduces the correlation between neighboring pixels to near zero, attains entropy values approaching the theoretical maximum of 8, and yields NPCR and UACI values close to their ideal benchmarks. Furthermore, it demonstrates strong resilience against cropping and noise attacks. These experimental findings confirm the algorithm’s superior performance in terms of cryptographic security, key sensitivity, and attack resistance, offering an effective and dependable approach for secure multi-image transmission.

Hyperspectral image (HSI) classification plays a vital role in advancing consumer-driven precision agriculture, particularly in complex coastal environments where spectral variability, salinity influence, and heterogeneous land–sea interactions pose significant analytical challenges. This study introduces HyperFusionMamba (HFMamba), an advanced coastal paradigm designed for robust and efficient HSI analysis in agricultural remote sensing. The proposed model integrates ScopeAttenConvolution (SAC), Lightweight Mixture of Experts (LMoE), SpatialNormAttention (SNA), and residual connections to achieve adaptive spectral–spatial fusion under coastal conditions. Specifically, SAC employs multi-scale dilated convolutions and channel attention to capture detailed spatial structures, LMoE enhances high-dimensional spectral feature learning through specialized expert kernels, and SNA strengthens spatial coherence and key-region representation in heterogeneous coastal scenes. Experimental results on benchmark HSI datasets demonstrate that HFMamba surpasses state-of-the-art models in Overall Accuracy (OA), Average Accuracy (AA), and Kappa coefficient (κ). This paradigm establishes a scalable and intelligent framework for coastal hyperspectral analysis, offering significant potential for consumer-driven precision agriculture and environmental observation.

This paper proposes a structurally simplified 2D quadratic sine map (2D-SQSM). This map effectively addresses the insufficient chaos performance of traditional chaotic maps while avoiding the overly complex structures of emerging chaotic maps. Evaluated using multiple chaos performance metrics, the 2D-SQSM demonstrates high Lyapunov exponents, and sample entropy, with chaotic characteristics superior to some advanced chaotic maps proposed in recent years. Based on the 2D-SQSM, this paper further designs a highly robust color image encryption algorithm. First, by introducing different hash functions multiple times, the correlation between the key and plaintext is enhanced, significantly improving resistance against brute-force attacks; second, cyclic shifting and segmentation-recombination operations are applied separately to the three RGB channels to effectively disrupt pixel distribution and significantly reduce spatial correlation between pixels; finally, the chaotic sequence generated by the 2D-SQSM is utilized for XOR diffusion, further enhancing the randomness and diffusion capability of the ciphertext. A large number of simulation results demonstrate that this algorithm can significantly enhance the image information entropy, and can effectively reduce pixel correlation, possessing good statistical properties. Furthermore, it is robust against differential attacks, noise attacks, cropping attacks, chosen plaintext attacks, etc., and is suitable for secure image transmission.

Mangrove ecosystems play a crucial role in coastal environments. However, due to the complexity of mangrove distribution and the similarity among different categories in remote sensing images, traditional image segmentation methods struggle to accurately identify mangrove regions. Deep learning techniques, particularly those based on CNNs and Transformers, have demonstrated significant progress in remote sensing image analysis. This study proposes TCCFNet (Two-Channel Cross-Fusion Network) to enhance the accuracy and robustness of mangrove remote sensing image semantic segmentation. This study introduces a dual-backbone network architecture that combines ResNet for fine-grained local feature extraction and Swin Transformer for global context modeling. ResNet improves the identification of small targets, while Swin Transformer enhances the segmentation of large-scale features. Additionally, a Cross Integration Module (CIM) is incorporated to strengthen multi-scale feature fusion and enhance adaptability to complex scenarios. The dataset consists of 230 high-resolution remote sensing images, with 80% used for training and 20% for validation. The experimental setup employs the Adam optimizer with an initial learning rate of 0.0001 and a total of 450 training iterations, using cross-entropy loss for optimization. Experimental results demonstrate that TCCFNet outperforms existing methods in mangrove remote sensing image segmentation. Compared with state-of-the-art models such as MSFANet and DC-Swin, TCCFNet achieves superior performance with a Mean Intersection over Union (MIoU) of 88.34%, Pixel Accuracy (PA) of 97.35%, and F1-score of 93.55%. Particularly, the segmentation accuracy for mangrove categories reaches 99.04%. Furthermore, TCCFNet excels in distinguishing similar categories, handling complex backgrounds, and improving boundary detection. TCCFNet demonstrates outstanding performance in mangrove remote sensing image segmentation, primarily due to its dual-backbone design and CIM module. However, the model still has limitations in computational efficiency and small-target recognition. Future research could focus on developing lightweight Transformer architectures, optimizing data augmentation strategies, and expanding the dataset to diverse remote sensing scenarios to further enhance generalization capabilities. This study presents a novel mangrove remote sensing image segmentation approach—TCCFNet. By integrating ResNet and Swin Transformer with the Cross Integration Module (CIM), the model significantly improves segmentation accuracy, particularly in distinguishing complex categories and large-scale targets. TCCFNet serves as a valuable tool for mangrove remote sensing monitoring, providing more precise data support for ecological conservation efforts.

This study introduces an innovative water depth estimation method for complex coastal environments, focusing on Yantian Port. By combining Random Forest algorithms with a Coordinate Attention mechanism, we address limitations of traditional bathymetric techniques in turbid waters. Our approach incorporates geographical coordinates, enhancing spatial accuracy and predictive capabilities of conventional models. The Random Forest Lon./Lat. model demonstrated exceptional performance, particularly in shallow water depth estimation, achieving superior accuracy metrics among all evaluated models. It boasted the lowest Root Mean Square Error (RMSE) and highest coefficient of determination (R²), outperforming standard techniques like Stumpf and Log-Linear approaches. These findings highlight the potential of advanced machine learning in revolutionizing bathymetric mapping for intricate coastal zones, opening new possibilities for port management, coastal engineering, and environmental monitoring of coastal ecosystems. We recommend extending this research to diverse coastal regions to validate its broader applicability. Additionally, exploring the integration of additional geospatial features could further refine the model’s accuracy and computational efficiency. This study marks a significant advancement in bathymetric technology, offering improved solutions for accurate water depth estimation in challenging aquatic environments. As we continue to push boundaries in this field, the potential for enhanced coastal management and environmental stewardship grows, paving the way for more sustainable and informed decision-making in coastal zones worldwide.