The dynamic environment of UAV swarms in forest management is characterized by communication instability, heterogeneous nodes, and frequent topology changes due to challenging terrain. These systems are vulnerable to network attacks, requiring advanced intrusion detection technologies. Traditional methods struggle with rapid changes due to data privacy concerns and centralized computational limits, while existing federated learning (FL) algorithms lack robustness against client heterogeneity and dynamic data distribution, especially in complex forest environments. To address these challenges, we propose federated intrusion detection for securing UAV swarms (FIDSUS). FIDSUS improves intrusion detection systems by leveraging collaborative sensing among UAVs, enabling better monitoring and response to security threats in forestry. By quantifying the similarity between UAVs’ local feature extractors through an affinity matrix, FIDSUS guides the aggregation of feature extractors, improving detection capabilities. It also uses AI-driven aerial and distributed computing to enhance data processing efficiency and decision-making speed. The framework addresses data heterogeneity by cross-round feature fusion, improving detection in dynamic environments. Experimental results on the NSL-KDD and UNSW-NB15 datasets show that FIDSUS outperforms existing FL methods with a 4%–34% accuracy improvement. FIDSUS shows robustness and accuracy in dynamic environments, providing an effective solution for securing UAV swarms in forestry.

The emergence of consumer electronic Unmanned Aerial Vehicle (UAV) systems has brought innovation to the field of ship inspection. Traditional ship detection is carried out through traditional target recognition methods, but the efficiency and accuracy cannot meet the requirements. With the major breakthrough in the resolution of remote sensing images, it has become possible to use UAV to capture remote sensing images to detect ships. In this paper, object detection technology based on deep learning is used to improve the current detection methods and achieve accurate ship target detection. We propose an anchor-free detection method based on FCOS to reduce model hyperparameters. Meanwhile, a positive and negative sample selection method is put forth based on attention mechanism feature fusion and self-adaptation to enhance the fusion expression of features and improve the efficiency of sample selection, therefore improving the accuracy of the model. Experiments demonstrate notable progress in the detection accuracy from the proposed method, especially when small ship targets are concerned. Compared with Faster R-CNN and R3Det, the method introduced in this paper needs fewer hyperparameters, while achieving higher detection accuracy, with AP50 reaching 83.90%.

The computational complexity of the Transformer model grows quadratically with input sequence length. This causes a sharp increase in computational cost and memory consumption for high-resolution remote sensing images. Consequently, its application in consumer-grade unmanned aerial vehicle remote sensing is limited. To address this issue, we propose an efficient machine learning-based semantic segmentation algorithm (EMLSSA). First, EMLSSA incorporates the hash clustering attention (HCAttention) mechanism. It employs the locality-sensitive hashing (LSH) algorithm to group similar features into hash buckets, enabling dynamic token clustering. Subsequently, tokens in the same hash bucket are aggregated by weighted summation. This compresses features and reduces the computational complexity of self-attention. Second, EMLSSA incorporates the frequency multi-layer perceptron (FMLP) mechanism. It combines frequency and spatial domain information, enhancing the ability of the Transformer to perceive local features. Experimental results show that EMLSSA-B4 reduces computational cost by 11.7% on FLAME, PWD, EarthVQA, and Potsdam datasets. Furthermore, it maintains comparable segmentation performance to SegFormer-B4.