In order to meet the demand for distributed joint timekeeping system, a remote time comparison system was de-signed, which includes satellite co-visibility, satellite bi-directional, and fiber bi-directional time comparison means. The domestic equipment used in the system has the same time comparison accuracy as imported equipment, achieving high-precision clock differ-ence measurement between laboratories using multiple means. The relative zero value calibration of the time comparison equipment was completed using a mobile calibration station, and a 20-day time comparison was conducted between laboratories that were 50 km apart. The experimental results show that the mean clock difference measured by the three comparison means at the two locations is less than 1 ns, the trend is the same, and the amplitude is similar, and all can accurately measure the clock difference between the two locations.

The complex electromagnetic environment of modern battlefields requires strong robustness of satellite navigation anti-jamming receivers. A polarization-sensitive array uses polarization information to increase the dimension of processing signals and can be applied in navigation to improve the anti-interference performance of the receiver. In this paper, the anti-jamming of the dual-polarization array in a complex electromagnetic environment is studied. The signal model and the dual-polarization array are in-troduced, and the amplitude and phase information obtained by finite element simulation on HFSS are then used to synthesize the steering vector and simulate it in MATLAB to prepare the anti-jamming ability of the ordinary array and the dual-polarization array. The real data is measured in a darkroom to verify the simulation results. The satellite collection positioning test was carried out on the roof to compare the positioning results of the ordinary array and the Dual-polarization array. The simulation and experiment shows that the dual-polarization array can suppress super-degree-of-freedom jamming and has good robustness under low elevation angle jamming. The dual-polarization antenna does not change the antenna size and can maintain the receiver miniaturization while keeping good robustness.

The chemical oxygen demand (COD) and chlorophyll a concentration, which are typical water quality parameters re-lated to the spectrum, serve as important indicators for reflecting the degree of water pollution and eutrophication. Support Vector Regression (SVR) is suitable for small sample sizes and widely utilized in remote sensing retrieval of typical offshore water quality parameters; however, it faces challenges in model parameter selection and may easily fall into local optimal solutions. To address this issue, an Improved Sparrow Search Algorithm (ISSA) is developed by integrating reverse learning and simulated annealing. An enhanced support vector regression model (ISA-SVR) is proposed by refining the Sparrow algorithm to optimize the penalty coeffi-cient and kernel parameters of the SVR model. Inversion models for COD and Chl-a concentrations are established using measured water spectra and data on water quality parameters. The accuracy of the model is validated using Sentinel-2 satellite remote sensing spectral data, yielding inversion accuracies for each water quality parameter concentration. The mean relative error (MRE) of the COD concentration prediction model and Chl-a concentration prediction model based on ISSA algorithm optimized SVR are 20.02%and 30.17%, respectively, outperforming other models such as linear regression, SVR, and SSA-SVR models. Experimental results demonstrate that ISA-SVR algorithm represents an effective approach for remotely sensed retrieval of COD and Chl-a concentrations while offering valuable insights for subsequent scientific management of offshore water quality.

The space-borne laser detection system has the capability to detect the vertical profile of clouds and aerosols. Exist-ing payloads, such as the Caliop radar on the US Calipso satellite, the multi-beam LiDAR on the domestic "Goumang" satellite, and the atmospheric detection lLiDAR on air pollution monitoring satellites, are limited to single-beam cloud-aerosol detection with a narrow detection area. A proposed solution is a multi-beam cloud-aerosol detection LiDAR system that operates in a satellite orbit of 800 km and utilizes a multi-beam detection system to extend radar coverage to 30 km. The central beam employs dual-wavelength polarization detection for obtaining vertical profiles and particle species of atmospheric aerosols and clouds, while the edge beam uses single-wavelength detection for cloud profiling. This approach significantly improves data acquisition efficiency through simulation-based verification of dynamic range and high sensitivity single photon detections which reduce laser energy require-ments, as well as radar weight and power consumption. Finally, through simulation, the detection effect of space-borne multi-beam cloud-aerosol detection radar on typical clouds and aerosols is verified.

By utilizing Unmanned Aerial Vehicle (UAV) Hyper-Spectral Imaging (HSI) and Light Detection and Ranging, this study aims to investigate the classification methods of wetland vegetation in the Yellow River estuary using LiDAR data. However,due to the high spatial resolution HSI spectral variability and uneven LiDAR point cloud density, the classification results exhibit a"pepper and salt" phenomenon. To address these issues, this paper proposes a two-branch convolutional neural network (SSF-C-DBCNN) that integrates empty spectrum feature fusion and channel attention mechanism. The spectral attention mechanism miti-gates the impact of spectral variability by assigning different weights to each band. Meanwhile, the spatial attention mechanism fo-cuses on learning and emphasizing dense point cloud regions with strong feature expression ability in order to alleviate the influence of uneven LiDAR point cloud density on the results. Finally, the channel attention mechanism is introduced for extracting deeper fea-tures after two-branch feature fusion. Experimental verification using HSI and LiDAR data collected by UAV demonstrates that the proposed method outperforms random forest as well as five deep learning methods, yielding more suitable classification results for actual land cover while effectively suppressing the "pepper and salt" phenomenon.

Lithological identification and classification constitute indispensable facets of geology, resource exploration, and re-lated disciplines. The emergence of hyperspectral remote sensing has ushered in novel perspectives for lithological identification. The utilization of machine learning to extract information from hyperspectral rock images, thereby enabling accurate lithological identification, holds paramount practical significance. Currently, the application of machine learning methods for the classification of hyperspectral rock images lacks a comprehensive exploitation of spatial and spectral information. Therefore, this paper introduces a three-dimensional convolutional residual network structure augmented with an attention mechanism, capable of effectively extracting spatial, spectral, and joint spatial-spectral features from hyperspectral rock images. In this experiment, images of 10 different types of rock samples were collected using a drone equipped with a hyperspectral sensor. The algorithm proposed in this study was applied to classify hyperspectral rock images. Experimental results indicate that, in comparison to traditional machine learning algo-rithms such as SVM and RF, as well as deep learning algorithms like ResNet, 3DCNN, and SSRN, the proposed algorithm exhibits higher accuracy.

Offshore ship object detection is a very challenging task and has received widespread attention from scholars and experts. Detectors based on Convolutional Neural Networks (CNN) and attention mechanisms have made significant progress in offshore ship object detection. However, the problem of false detection in the detection process is caused by the apparent similarity and background interference of ship targets. In order to solve this problem, this paper proposes a detection head module for fine-grained appearance discrimination implemented with Faster RCNN. This module includes a category fine-grained branch and an efficient full-dimensional dynamic convolution localization branch. The category fine-grained branch mines and utilizes category fine-grained identification features through global feature modeling and flexible perception range. The efficient omni-dimensional dynamic convolution positioning branch distinguishes objects and backgrounds through the efficient and flexible perception of ship boundary information, thereby reducing false and missed detections. Through experimental verification on the offshore ship public dataset Seaships7000, the proposed algorithm reduces false detections and missed detections and improves detector performance.

Hyperspectral target detection is crucial in Earth observation for both military and civilian applications. However,complex backgrounds and the scarcity of target samples pose challenges in hyperspectral image analysis. In this paper, we first employ the CEM coarse detection method to extract background data. Subsequently, a novel knowledge distillation model, namely KDTGAN (implemented through Transformer-GAN), is introduced. The generator of this teacher model adopts the structure of a Transformer encoder and combines it with a multi-scale data fusion approach to accurately learn the background distribution, which in turn enables target detection by reconstructing the background information. To overcome the challenge of unstable GAN training,especially the scarcity of pure background data, we propose a new loss algorithm to reduce the negative impact of suspicious target samples on model performance. To reduce the computational burden of the model, we introduce knowledge distillation and design a new distillation loss to constrain the student model to lighten the model while improving the student model's detection accuracy. The experimental results show that KDTGAN performs better than current detection methods with higher detection accuracy and robustness.

With the rapid development of remote sensing technology, the intelligent decoding of weak targets in optical remote sensing images has become one of the research hotspots in remote sensing information processing. The feature targets of remote sensing images are often characterized by small scale, many types, a large number, fast moving speed of some key small targets, and are easily affected by the complex background environment and noise, which makes it a great challenge extract information from weak targets in remote sensing images. Early research on weak target segmentation, detection, and tracking algorithms in intelligent interpretation algorithms mostly relied on template matching and a priori knowledge, and such algorithms need to consume a lot of resources, arithmetic, and expert knowledge costs, and there were problems of large computational volume and poor generalization ability. In recent years, with the rapid development of deep learning and other artificial intelligence technologies, the information of weak targets can be accurately obtained in massive remote sensing data, and the features of weak targets can be quickly extracted by combining deep learning algorithms to provide efficient and accurate decoding information. This paper summarizes the research progress of intelligent interpretation algorithms for weak targets in remote sensing images, including weak target segmentation, detection, and tracking algorithms based on traditional image processing methods, as well as typical related algorithms based on deep learning. By analyzing the advantages and limitations of these methods, it is of great significance to improve the information acquisition ability of relevant targets, enhance the situational awareness level of observation, and future applications.

As a difficulty and focus in the field of radar imaging, radar forward-looking imaging has broad application prospects in automatic driving, navigation, precision guidance and so on. The traditional forward-looking imaging algorithm is limited by the width of the antenna aperture and cannot achieve high-resolution imaging. In this paper, CNN ( Convolutional Neural Networks )neural network and LSTM ( Long Short-Term Memory ) neural network are combined to realize the prediction of azimuth in forward-looking imaging. Firstly, the convolution-like model of the scanning forward-looking imaging signal and its ill-posedness are introduced. The echo signal is preprocessed by pulse compression and range migration correction, and input into the CNN-LSTM neural network to perform azimuth estimation by range unit. The simulation results show that the algorithm can effectively improve the azimuth resolution of forward-looking imaging and realize the super-resolution of forward-looking imaging.