The detection of train obstacle intrusion is very important for the safe running of trains. In this paper, we design a multitask intrusion detection model to warn of the intrusion of detected target obstacles in railway scenes. In addition, we design a multiobjective optimization algorithm that performs with different task complexity. Through the shared structure reparameterized backbone network, our multitask learning model utilizes resources effectively. Our work achieves competitive results on both object detection and line detection, and achieves excellent inference time performance (50 FPS). Our work is the first to introduce a multitask approach to realize the assisted-driving function in a railway scene.

Underwater video object detection is a challenging task due to the poor quality of underwater videos, including blurriness and low contrast. In recent years, Yolo series models have been widely applied to underwater video object detection. However, these models perform poorly for blurry and low-contrast underwater videos. Additionally, they fail to account for the contextual relationships between the frame-level results. To address these challenges, we propose a video object detection model named UWV-Yolox. First, the Contrast Limited Adaptive Histogram Equalization method is used to augment the underwater videos. Then, a new CSP_CA module is proposed by adding Coordinate Attention to the backbone of the model to augment the representations of objects of interest. Next, a new loss function is proposed, including regression and jitter loss. Finally, a frame-level optimization module is proposed to optimize the detection results by utilizing the relationship between neighboring frames in videos, improving the video detection performance. To evaluate the performance of our model, We construct experiments on the UVODD dataset built in the paper, and select mAP@0.5 as the evaluation metric. The mAP@0.5 of the UWV-Yolox model reaches 89.0%, which is 3.2% better than the original Yolox model. Furthermore, compared with other object detection models, the UWV-Yolox model has more stable predictions for objects, and our improvements can be flexibly applied to other models.

Uterine myomas affect 70% of women of reproductive age, potentially impacting their fertility and health. Manual film reading is commonly used to identify uterine myomas, but it is time-consuming, laborious, and subjective. Clinical treatment requires the consideration of the positional relationship among the uterine wall, uterine cavity, and uterine myomas. However, due to their complex and variable shapes, the low contrast of adjacent tissues or organs, and indistinguishable edges, accurately identifying them in MRI is difficult. Our work addresses these challenges by proposing an instance segmentation network capable of automatically outputting the location, category, and masks of each organ and lesion. Specifically, we designed a new backbone that facilitates learning the shape features of object diversity, and filters out background noise interference. We optimized the anchor box generation strategy to provide better priors in order to enhance the process of bounding box prediction and regression. An adaptive iterative subdivision strategy ensures that the mask boundary details of objects are more realistic and accurate. We conducted extensive experiments to validate our network, which achieved better average precision (AP) results than those of state-of-the-art instance segmentation models. Compared to the baseline network, our model improved AP on the uterine wall, uterine cavity, and myomas by 8.8%, 8.4%, and 3.2%, respectively. Our work is the first to realize multiclass instance segmentation in uterine MRI, providing a convenient and objective reference for the clinical development of appropriate surgical plans, and has significant value in improving diagnostic efficiency and realizing the automatic auxiliary diagnosis of uterine myomas.

Medical image segmentation can effectively identify lesions in medicine, but some small and rare lesions cannot be well identified. Existing studies do not take into account the uncertainty of the occurrence of diseased tissue, and the problem of long-tailed distribution of medical data. Meanwhile, the grayscale image obtained from Magnetic Resonance Imaging (MRI) detection has problems, such as the features being difficult to extract and invalid features being difficult to distinguish. In order to solve these problems, we propose a new weighted attention ResUNet (WA-ResUNet) and a class weight formula based on the number of images contained in the class, which improves the performance of the model in the low-frequency class and the overall effect of the model by improving the degree of attention paid to the valid features and invalid ones and rebalancing the learning efficiency among the classes. We evaluated our method on an uterine MRI dataset and compared it with the ResUNet. WA-ResUNet increased Intersection over Union (IoU) in the low-frequency class (Nabothian cysts) by 21.87%, and the overall mIoU increased by more than 6.5%.

Adversarial examples pose a significant threat to Deep Neural Networks (DNNs) underpinning Synthetic Aperture Radar (SAR) Automatic Target Recognition (ATR) systems, as these models exhibit acute susceptibility to such malicious inputs. While white-box attacks achieve high success rates, their transferability to unknown black-box models—particularly across different network architectures (e.g., from CNNs to Vision Transformers)—remains a significant challenge. Existing gradient-based iterative methods often overfit the specific decision boundary of the surrogate model, resulting in poor generalization. To address this, we propose a novel generative attack framework termed BUM. Instead of merely maximizing the classification error, BUM explicitly models and minimizes the epistemic uncertainty of the surrogate model. By leveraging Monte Carlo (MC) Dropout to simulate a Bayesian ensemble, we train a generator to craft perturbations that are consistently adversarial across stochastic sub-models. This regularization forces the attack to target high-level, structure-aware semantic features shared among architectures, rather than low-level, model-specific artifacts. Extensive experiments on the MSTAR and FUSAR datasets demonstrate the superior black-box transferability of BUM.