Xin Nie

Analyzing flight data from high-altitude unmanned aerial vehicles (UAVs) is an essential approach to anomaly detection and early warning. However, flight data collected during the climb, cruise, and descent phases of high-altitude UAV missions exhibit pronounced distributional differences, making it difficult for existing anomaly detection methods to establish a unified model of normal flight behavior across all phases, thereby constraining detection accuracy. To address this, this paper proposes an anomaly detection method for high-altitude UAV flight data based on a Mixture of Experts (MoE). Specifically, a Multi-Scale Dual-Path Mixture-of-Experts encoder is constructed to extract local perturbations from UAV flight sequences; multi-scale convolution, a Transformer, and a dual-path LSTM are fused to capture temporal dependencies within the flight data; and a gating network automatically adapts to and uniformly models the data distributions across different mission phases. Furthermore, a two-stage reconstruction error is employed to construct sliding-window anomaly scores, and the median absolute deviation (MAD) is adopted to adaptively determine the anomaly decision threshold, enabling detection of flight-data anomalies across all mission phases. Experimental results on a self-constructed high-altitude UAV flight dataset with controlled anomaly injections demonstrate that the proposed method achieves a window F1 of 93.97% and an event recall of 91.77% across three independent training seeds.

Analyzing flight data from high-altitude unmanned aerial vehicles (UAVs) is an essential approach to fault detection and early warning. However, flight data collected during the climb, cruise, and descent phases of high-altitude UAV missions exhibit pronounced distributional differences, making it difficult for existing anomaly detection methods to establish a unified fault model across all phases, thereby constraining detection accuracy. To address this, this paper proposes an anomaly detection method for high-altitude UAV flight data based on a Mixture of Experts (MoE). Specifically, a Multi-Scale Dual-Path Mixture of Experts encoder is constructed to extract local perturbations from UAV flight sequences; multi-scale convolution, a Transformer, and a dual-path LSTM are fused to capture temporal dependencies within the flight data; and a gating network automatically adapts to and uniformly models the data distributions across different mission phases. Furthermore, a two-stage reconstruction error is employed to construct sliding-window anomaly scores, and the Median Absolute Deviation (MAD) is adopted to adaptively determine the anomaly decision threshold, enabling accurate detection of flight-data anomalies across all mission phases. Experimental results on a self-constructed high-altitude UAV flight dataset demonstrate that the proposed method achieves an accuracy, recall, and F1-score of 95.69%, 92.35%, and 93.79%, respectively, providing effective support for anomaly early warning over multi-phase flight data.

Infrared and visible image fusion combines the thermal cues captured by infrared sensors with the rich structural and texture information provided by visible images. This technique is particularly valuable for security surveillance, nighttime scene perception, and target recognition under challenging environmental conditions. Existing methods generally adopt fixed fusion strategies and neglect dynamic dual-modality changes under low illumination, overexposure and strong light interference, failing to preserve both thermal target saliency and visible structural details in fused images. To address this problem, this paper proposes a Lighting-Aware Spatial–Frequency Fusion Network (LASFNet) for security surveillance. It first estimates modality reliability across low-light, overexposed and infrared-salient regions, incorporating it into the fusion of frequency-domain amplitude and phase. Spatial infrared, visible and frequency-domain compensation features are then jointly fused to generate the output. Experiments on M3FD, MSRS and RoadScene datasets show that LASFNet achieves competitive fusion performance with strong cross-dataset generalization. On M3FD, YOLOv8s with LASFNet-fused inputs achieves 84.825% mAP@0.5 and 57.319% mAP@0.5:0.95, outperforming visible, infrared and YDTR-fused inputs. The proposed method balances target saliency and scene structure, providing more effective visual input for object detection under complex illumination.

Object detection of small and occluded targets in complex scenarios is a vital yet challenging task in computer vision, with applications in intelligent systems (e.g., kitchen safety supervision). To address limitations of existing models, this study proposes an improved YOLOv10 algorithm with three key innovations. We first introduce a Mosaic-9 data augmentation strategy to enhance small target density in training samples. The traditional PANet in YOLOv10 is replaced by the Bidirectional Feature Pyramid Network (BiFPN), which uses cross-scale bidirectional connections and learnable weights to optimize multi-scale feature fusion. A Squeeze-and-Excitation (SE) channel attention module is integrated into the CSPDarknet backbone to emphasize key feature channels and mitigate background interference for occluded objects. Experiments on a self-constructed dataset (6508 images with multi-scale and occluded targets) show the improved YOLOv10 achieves 69.5% mAP@0.5, a 7.7 percentage point increase over YOLOv10n, while maintaining 12.1 ms inference speed. Ablation studies verify Mosaic-9 enhances small target perception, BiFPN boosts mAP@0.5 by 5.7%, and SE improves occlusion robustness by 4.8%. This work offers a generalizable multi-module optimization framework for YOLO-series models, applicable to various small and occluded target detection tasks, advancing lightweight object detection algorithms and intelligent vision systems.

Polycystic Ovary Syndrome (PCOS) is a widespread condition in women that is associated with hormonal disorders, absentee or infrequent menstruation, and cyst formation on ovaries. The diagnostic process of PCOS is still difficult because of ambiguity in ultrasound images. Earlier approaches emphasize the detection and classification of follicles for ultrasound images and disregard interpretability, which obscures understanding the model’s decisions and trustworthiness in critical medical applications. This research proposes a method of diagnosing PCOS in two stages. Initially, follicles are located with a YOLOv8 model that has multiscale feature dose attention. Follicles are subsequently classified with a ResNet50 model that has SE blocks. The last step is for Grad-CAM to show which features of the image were used by the classification model in order to explain its decision and provide meaningful insights regarding the model's predictions. Evaluation was performed on two publicly available datasets. The proposed method outperforms all other methods in follicle detection with 92% mean average precision (mAP) and classification accuracies of 98.25% and 94.56% on datasets 1 and 2, respectively. This result makes the proposed model a reliable and transparent technique for robust clinical applications. By providing a scalable, deployable, and interpretable diagnostic pipeline that can be integrated into AI-based health-tech platforms, the suggested method also fits with the objectives of technopreneurship incubator models.

Polycystic Ovary Syndrome (PCOS) is a common endocrine disorder in women that is frequently diagnosed by ultrasound imaging by examining ovarian abnormality. Ultrasound images, though, are normally contaminated with speckle noise that degrades image quality and poses difficulties for diagnosis. Conventional denoising methods like mean and median filtering are unable to eliminate noise properly while maintaining fine details. To solve this problem, this paper introduces a new Attention-based Autoencoder (AAE) for denoising PCOS ultrasound images. The model uses an attention mechanism to selectively amplify significant image areas and suppress noise, enhancing image quality for diagnosis. The introduced method was tested on a publicly available ultrasound dataset with synthetic speckle noise at various levels (variance 0.5, 0.02, and 0.001). Experimental results prove that the suggested method performs better than conventional denoising methods, with peak signal-to-noise ratio (PSNR) values of 31.33, 34.25, and 36.23, respectively. The structural similarity index measure (SSIM) also reveals notable improvements and corresponding scores of 85.21, 92.33, and 99.25. Beyond technical performance, this work supports the development of scalable, AI-driven PCOS diagnostic tools within a technopreneurship incubator model. These results indicate that AAEs can improve ultrasound image quality, facilitating more accurate PCOS diagnosis.

To address the mission planning challenge for agile satellites in dense point target observation, a clustering strategy based on an ant colony algorithm and a heuristic simulated genetic annealing optimization algorithm are proposed. First, the imaging observation process of agile satellites is analyzed, and an improved ant colony algorithm is employed to optimize the clustering of observation tasks, enabling the satellites to complete more observation tasks efficiently with a more stable attitude. Second, to solve for the optimal group target observation sequence and achieve higher total observation benefits, a task planning model based on multi-target observation benefits and attitude maneuver energy consumption is established, considering the visible time windows of targets and the time constraints between adjacent targets. To overcome the drawbacks of traditional simulated annealing and genetic algorithms, which are prone to local optimal solution and a slow convergence speed, a novel Simulated Genetic Annealing Algorithm is designed while optimizing the sum of target observation weights and yaw angles while also accounting for factors such as target visibility windows and satellite attitude transition times between targets. Ultimately, the feasibility and efficiency of the proposed algorithm are substantiated by comparing its performance against traditional heuristic optimization algorithms using a dataset comprising large-scale dense ground targets.

In order to solve the existing distracted driving behaviour detection algorithms’ problems such as low recognition accuracy, high leakage rate, high false recognition rate, poor real-time performance, etc., and to achieve high-precision real-time detection of common distracted driving behaviours (mobile phone use, smoking, drinking), this paper proposes a driver distracted driving behaviour recognition algorithm based on YOLOv5. Firstly, to address the problem of poor real-time identification, the computational and parametric quantities of the network are reduced by introducing a lightweight network, Ghostnet. Secondly, the use of GSConv reduces the complexity of the algorithm and ensures that there is a balance between the recognition speed and accuracy of the algorithm. Then, for the problem of missed and misidentified cigarettes during the detection process, the Soft-NMS algorithm is used to reduce the problems of missed and false detection of cigarettes without changing the computational complexity. Finally, in order to better detect the target of interest, the CBAM is utilised to enhance the algorithm’s attention to the target of interest. The experiments show that on the homemade distracted driving behaviour dataset, the improved YOLOv5 model improves the mAP@0.5 of the YOLOv5s by 1.5 percentage points, while the computational volume is reduced by 7.6 GFLOPs, which improves the accuracy of distracted driving behaviour recognition and ensures the real-time performance of the detection speed.