Maritime monitoring has become a critical issue for both industry and academia, with the dual goals of ensuring safety and mitigating environmental impacts. This paper proposes a semantic communication system tailored for maritime monitoring, which is based on the Swin transformer-based Deep Semantic Autoencoder (Swin-DSA). The proposed Swin-DSA consists of the camera, encoder, decoder, and classifier. The camera, mounted on an unmanned aerial vehicle, captures images of the maritime monitoring area, whereas the encoder employs swin transformer to extract and encode semantic information from the images, which are then transmitted through maritime wireless communication channels to the shore-based command center. The decoder decodes the signals and restores the semantic information utilizing the swin transformer. Furthermore, the classifier employs the GoogleNet V2 to recognize and classify the decoded images. The proposed Swin-DSA optimizes its network parameters by minimizing both the reconstruction loss and the classification loss. Experimental results demonstrate that the proposed Swin-DSA not only achieves the extraction, transmission, restoration, and reconstrution of image semantics, but also outperforms the benchmark in terms of peak signal-to-noise ratio and classification accuracy. Therefore, the proposed Swin-DSA provides a new technical reference for maritime monitoring.

With the rapid growth of the global marine economy and flourishing maritime activities, the marine Internet of Things (IoT) is gaining unprecedented momentum. However, current marine equipment is deficient in data transmission efficiency and semantic comprehension. To address these issues, this paper proposes a novel End-to-End (E2E) coding scheme, namely the Turbo-based Deep Semantic Autoencoder (Turbo-DSA). The Turbo-DSA achieves joint source-channel coding at the semantic level through the E2E design of transmitter and receiver, while learning to adapt to environment changes. The semantic encoder and decoder are composed of transformer technology, which efficiently converts messages into semantic vectors. These vectors are dynamically adjusted during neural network training according to channel characteristics and background knowledge base. The Turbo structure further enhances the semantic vectors. Specifically, the channel encoder utilizes Turbo structure to separate semantic vectors, ensuring precise transmission of meaning, while the channel decoder employs Turbo iterative decoding to optimize the representation of semantic vectors. This deep integration of the transformer and Turbo structure is ensured by the design of the objective function, semantic extraction, and the entire training process. Compared with traditional Turbo coding techniques, the Turbo-DSA shows a faster convergence speed, thanks to its efficient processing of semantic vectors. Simulation results demonstrate that the Turbo-DSA surpasses existing benchmarks in key performance indicators, such as bilingual evaluation understudy scores and sentence similarity. This is particularly evident under low signal-to-noise ratio conditions, where it shows superior text semantic transmission efficiency and adaptability to variable marine channel environments.

Aiming at the problems of high complexity and low detection accuracy of single-stage three-dimensional (3D) detection method, a vehicle object detection algorithm based on the Efficient Channel Attention (ECA) mechanism is proposed. This paper provides a good solution to the problems of low object recognition accuracy and high model complexity in the field of 3D object detection. First, we voxelized the original point cloud data, taking the average coordinates and intensity values as the initial features. By entering into the Voxel Feature Encoding (VFE) layer, we can extract the features of each voxel. Then, referring to the VoxelNet model, the ECA mechanism is introduced, which reduces the complexity of the model while maintaining the good performance in the model. Finally, experiments on the widely used KITTI dataset show that the algorithm performs well, and the accuracy of the proposed ECA algorithm has reached 87.75%. Compared with the current mainstream algorithm SE-SSD of object detection, the accuracy is increased by 0.21%.

Since the BP neural network has poor performance and unstable learning rate in the maximum power point tracking (MPPT) algorithm of photovoltaic (PV) system, an adaptive particle swarm optimization BP neural network-fuzzy control PV MPPT algorithm (APSO-BP-FLC) is proposed in this paper. First, the inertia weight, learning factor and acceleration factor of particle swarm optimization (PSO) are self-updating, and the mutation operator is adopted to initialize the position of each particle. Second, the APSO algorithm is used to update the optimal weight threshold of BP neural network, where the input layer is irradiation and temperature, and the output layer is the maximum power point (MPP) voltage. Third, the fuzzy logical control (FLC) is employed to adjust the duty cycle of Boost converter. The inputs of FLC are voltage difference and duty ratio D(n-1) at the previous time, and the output is duty ratio D(n). Moreover, D(n-1) is optimized by |dP/dU| to improve the search range of FLC. The irradiation, temperature and MPP voltage of PV cell are adopted as the datasets for simulation in a city in Shaanxi province, China. Simulation results show that the proposed MPPT algorithm is superior to the APSO-BP, FLC and perturbation and observation (P&O) algorithm with tracking performance, steady state oscillation rate and efficiency. In addition, the efficiency of proposed MPPT algorithm is improved by 0.37%, 6.2%, and 6.8% as compared to APSO-BP, FLC and P&O algorithm.

针对游戏内玩家付费情况预测的问题，通过对数据集的清洗、降维以及数据分析，对游戏内玩家ID、在线时长以及7 d内的付费情况等特征数据的提取和分析，并预测玩家前45 d的付费情况，通过4种算法的对比分析得出Light GBM模型的预测结果更加准确、效率更高。

Aiming at the problems of inaccurate recommendation and single consideration in the traditional Points of Interest (POI) recommendation model, a POI Recommendation System using Hypergraph Embedding and Logical Matrix Factorization (HE-LMF) has been proposed. The user's check-in points of interest and time information are sampled by hypergraph embedding technology, and users with similar points of interest to the target user are found, and their points of interest are recommended to the target user. At the same time, through the geographic recommendation model based on logical matrix decomposition, the regions with many user check-in times and the correlation of each region are considered. The results of the two models are weighted, and top-k is selected to recommend to the user. Finally, experiments are carried out on the two datasets of gowalla and foursquare, and compared with the three models USG, PFMMGM and LRT. The experimental results show that the HE-LMF algorithm can effectively improve the accuracy and recall rate of POI recommendation.

Most of the traditional recommendation algorithm models are recommended based on the user's own historical preferences, although it can recommend POI for users to a certain extent. But in real life, people are more willing to ask their friends what they think when they have a difficult decision. Therefore, a word2vec-based social relationship point of interest recommendation model (W-SimTru) is proposed, which combines the similarity of friends based on cosine similarity with the friend trust recommendation algorithm based on TF-IDF to improve the model recommendation effect. In addition, before modeling the similarity of users, word2vec is used to process the user's historical check-in behavior to solve the problem of inaccurate recommendation due to sparse check-in data. Finally, experiments are carried out on three datasets of Los Angeles, Washington and NYC in Gowalla, and the experimental results show that the proposed W-SimTru recommendation algorithm outperforms the algorithms of the three comparative experiments.

Channel estimation is a key part of communication systems. However, the channel of millimeter-Wave (mmWave) Massive Multiple-Input Multiple-Output (Massive-MIMO) system has sparse characteristics, and the conventional channel estimation method is prone to noise factors and tends to achieve low estimation accuracy. Therefore, in this paper a channel estimation method is proposed for mmWave Massive MIMO systems based on deep learning. Firstly, a dataset to simulate a real-world environment, is generated by setting specific parameters. Furthermore, the generated channel matrix is adopted as the input of the neural network. Secondly, the attention mechanism is integrated into the deep learning method with ResUNet to enhance the ability of feature extraction. Finally, the channel estimation matrix is obtained via the aforementioned network model. The experimental results indicate that the Massive-MIMO method is superior to the conventional channel estimation methods in channel estimation accuracy and convergence rate, and has a good application prospects.