With the rapid advancement of remote-sensing technology, the spectral information obtained from hyperspectral remote-sensing imagery has become increasingly rich, facilitating detailed spectral analysis of Earth’s surface objects. However, the abundance of spectral information presents certain challenges for data processing, such as the “curse of dimensionality” leading to the “Hughes phenomenon”, “strong correlation” due to high resolution, and “nonlinear characteristics” caused by varying surface reflectances. Consequently, dimensionality reduction of hyperspectral data emerges as a critical task. This paper begins by elucidating the principles and processes of hyperspectral image dimensionality reduction based on manifold theory and learning methods, in light of the nonlinear structures and features present in hyperspectral remote-sensing data, and formulates a dimensionality reduction process based on manifold learning. Subsequently, this study explores the capabilities of feature extraction and low-dimensional embedding for hyperspectral imagery using manifold learning approaches, including principal components analysis (PCA), multidimensional scaling (MDS), and linear discriminant analysis (LDA) for linear methods; and isometric mapping (Isomap), locally linear embedding (LLE), Laplacian eigenmaps (LE), Hessian locally linear embedding (HLLE), local tangent space alignment (LTSA), and maximum variance unfolding (MVU) for nonlinear methods, based on the Indian Pines hyperspectral dataset and Pavia University dataset. Furthermore, the paper investigates the optimal neighborhood computation time and overall algorithm runtime for feature extraction in hyperspectral imagery, varying by the choice of neighborhood k and intrinsic dimensionality d values across different manifold learning methods. Based on the outcomes of feature extraction, the study examines the classification experiments of various manifold learning methods, comparing and analyzing the variations in classification accuracy and Kappa coefficient with different selections of neighborhood k and intrinsic dimensionality d values. Building on this, the impact of selecting different bandwidths t for the Gaussian kernel in the LE method and different Lagrange multipliers λ for the MVU method on classification accuracy, given varying choices of neighborhood k and intrinsic dimensionality d, is explored. Through these experiments, the paper investigates the capability and effectiveness of different manifold learning methods in feature extraction and dimensionality reduction within hyperspectral imagery, as influenced by the selection of neighborhood k and intrinsic dimensionality d values, identifying the optimal neighborhood k and intrinsic dimensionality d value for each method. A comparison of classification accuracies reveals that the LTSA method yields superior classification results compared to other manifold learning approaches. The study demonstrates the advantages of manifold learning methods in processing hyperspectral image data, providing an experimental reference for subsequent research on hyperspectral image dimensionality reduction using manifold learning methods.

Overstocked dumping sites associated with open-pit coal mining occupy original vegetation areas and cause damage to the environment. The monitoring of vegetation disturbance and restoration at dumping sites is important for the accurate planning of ecological restoration in mining areas. This paper aimed to monitor and assess vegetation disturbance and restoration in the dumping sites of the Baorixile open-pit mine using the LandTrendr algorithm and remote sensing images. Firstly, based on the temporal datasets of Landsat from 1990 to 2021, the boundaries of the dumping sites in the Baorixile open-pit mine in Hulunbuir city were extracted. Secondly, the LandTrendr algorithm was used to identify the initial time and duration of vegetation disturbance and restoration, while the Normalized Difference Vegetation Index (NDVI) was used as the input parameter for the LandTrendr algorithm. Thirdly, the vegetation restoration effect at the dumping sites was monitored and analyzed from both temporal and spatial perspectives. The results showed that the dumping sites of the Baorixile open-pit mine were disturbed sharply by the mining activities. The North dumping site, the South dumping site, and the East dumping site (hereinafter referred to as the North site, the South site, and the East site) were established in 1999, 2006, and 2010, respectively. The restored areas were mainly concentrated in the South site, the East site, and the northwest of the North site. The average restoration intensity in the North site, South site, and East site was 0.515, 0.489, and 0.451, respectively, and the average disturbance intensity was 0.371, 0.398, and 0.320, respectively. The average restoration intensity in the three dumping sites was greater than the average disturbance intensity. This study demonstrates that the combination of temporal remote sensing images and the LandTrendr algorithm can follow the vegetation restoration process of an open-pit mine clearly and can be used to monitor the progress and quality of ecological restoration projects such as vegetation restoration in mining areas. It provides important data and support for accurate ecological restoration in mining areas.

Automatic coordinated control of power transformers is essential to stable operation of power systems. However, there still lack mature intelligent solutions for this purpose. As a result, this paper proposes a reinforcement learning-based automatic collaborative control approach for power transformers. Firstly, by establishing the state space and action of the transformer control system, deep reinforcement learning is used to optimize the control strategy. Then, deep Q-network is utilized to automatically learning and adjust the operating parameters of each transformer to achieve coordinated control of power systems. In the algorithm design, we consider multiple factors such as power grid load and voltage stability. And they are incorporated into the reward function of reinforcement learning to facilitate learning appropriate control strategies. The experimental results show that the new method not only improves the response speed of the control system, but also effectively enhances the stability and robustness. In addition, we conducted in-depth analysis on the convergence and learning efficiency of the system, verifying the feasibility and superiority of the proposed method compared with traditional methods.

With the increasing importance of forest resources management, accurate classification of forest tree species has become a key task. Based on airborne LiDAR (Light Detection and Ranging) and hyperspectral technology, a comprehensive application system is proposed to improve the accuracy of automatic classification of forest tree species. In this study, the airborne LiDAR data, and AISA Eagle Ⅱ hyperspectral data of a natural forest in mountainous area of a forest farm were taken as the objects, and a combined system of airborne LIDAR and hyperspectral data was established to study the classification of main forest species in mountainous areas. In this experiment, the image obtained by PCA (Principal Component Analysis) transformation and DCHM (Digital canopy height model) data are combined into a fused image by band fusion. In the experiment, RF (Random Forest) method, one of the supervised classification methods, was selected to classify forest tree species from the extracted single-point feature vectors. The findings of the study indicate that, in this experiment, the classification accuracy achieved through fusion surpasses that achieved solely through hyperspectral data and airborne LiDAR. The overall accuracy sees an improvement of 28.8% and 15.6% compared to the individual modalities, accompanied by respective increases in the Kappa coefficient by 0.407 and 0.184. Consequently, the classification performance is deemed optimal, and the level of recognition is notably elevated. When only hyperspectral data is used as the classification basis, there are more misclassification points, but the characteristics of point cloud are improved to some extent. The system of integrated application of airborne LiDAR and hyperspectral technology provides an efficient and accurate solution for forest tree species classification. This method can play an important role in forest resources management, ecological monitoring and other fields, and provide strong support for the protection and sustainable utilization of forest resources.