Adding spatial penalty to fuzzy C-means (FCM) model is an important way to reduce the influence of noise in image segmentation. However, these improved algorithms easily cause segmentation failures when the image has the characteristics of unequal cluster sizes. Besides, they often fall into local optimal solutions if the initial cluster centers are improper. This paper presents a noise robust hybrid algorithm for segmenting image with unequal cluster sizes based on chaotic crow search algorithm and improved fuzzy c-means to overcome the above defects. Firstly, each size of clusters is integrated into the objective function of noise detecting fuzzy c-means algorithm (NDFCM), which can reduces the contribution of larger clusters to objective function and then the new membership degree and cluster centers are deduced. Secondly, a new expression called compactness, representing the pixel distribution of each cluster, is introduced into the iteration process of clustering. Thirdly, we use two- paths to seek the optimal solutions in each step of iteration: one path is produced by the chaotic crow search algorithm and the other is originated by gradient method. Furthermore, the better solutions of the two-paths go to next generation until the end of the iteration. Finally, the experiments on the synthetic and non–destructive testing (NDT) images show that the proposed algorithm behaves well in noise robustness and segmentation performance.
Human action recognition has received much research attention in recent years. Discriminative features learning is important for modeling the spatial and temporal evolutions of different actions. Recent works attempted to use RNN/LSTM to learn spatio-temporal features of joints. However, the features of joints are not fully explored in these conventional LSTMs, where each joint only receives context information from itself from previous frames. In this paper, we propose a LSTM network model by using aggregated feature learning strategy to extend feature representation in both spatial and temporal domains. In this model, the features of a joint are aggregated not only from all joints at previous moments, but also at current moment. Therefore, action-related features within the human skeleton sequences can be better represented and analyzed. To further optimize the network, we present a reparameterization that involves batch normalization and sparsemax classifier for the proposed LSTM network. We evaluate our approach on NTU RGB+D benchmark dataset. Experiment results demonstrate the effectiveness of the proposed approaches.
Accurate identification of the endometrial region is critical for the early detection of endometrial lesions. However, current detection models still face two major challenges when processing endometrial imaging data: (1) In complex and noisy environments, recognition accuracy remains limited, partly due to the insufficient exploitation of color information within the images; (2) Traditional Two-dimensional PCA-based (2DPCA-based) feature selection methods have limited capacity to capture and represent key characteristics of the endometrial region. To address these challenges, this paper proposes a novel algorithm named Feature-Level Image Fusion and Improved Swarm Intelligence Optimization Algorithm (FLFSI), which integrates a learning guided binary particle swarm optimization (BPSO) strategy with an image feature selection and reconstruction framework to enhance the detection of endometrial regions in clinical ultrasound images. Specifically, FLFSI contributes to improving feature selection accuracy and image reconstruction quality, thereby enhancing the overall performance of region recognition tasks. First, we enhance endometrial image representation by incorporating feature engineering techniques that combine structural and color information, thereby improving reconstruction quality and emphasizing critical regional features. Second, the BPSO algorithm is introduced into the feature selection stage, improving the accuracy of feature selection and its global search ability while effectively reducing the impact of redundant features. Furthermore, we refined the BPSO design to accelerate convergence and enhance optimization efficiency during the selection process. The proposed FLFSI algorithm can be integrated into mainstream detection models such as YOLO11 and YOLOv12. When applied to YOLO11, FLFSI achieves 96.6% Box mAP and 87.8% Mask mAP. With YOLOv12, it further improves the Mask mAP to 88.8%, demonstrating excellent cross-model adaptability and robust detection performance. Extensive experimental results validate the effectiveness and broad applicability of FLFSI in enhancing endometrial region detection for clinical ultrasound image analysis.
Based on the principle of color (RGB) filtering, an improved adaptive video caption detection and extraction method is proposed. Firstly, the principle of the color filtering algorithm used in the video caption detection and extraction method is analyzed, and then the algorithm is improved adaptively according to the caption pixel size to filter the noise. Finally, experiments verify the effect of this method in extracting subtitles from video. The experimental results show that the accuracy of detecting and extracting subtitles in color video is as high as 99.3.
Endometrial lesions are common in gynecology, exhibiting considerable clinical heterogeneity across different subtypes. Although ultrasound imaging is the preferred diagnostic modality due to its noninvasive, accessible, and cost-effective nature, its diagnostic performance remains highly operator-dependent, leading to subjectivity and inconsistent results. To address these limitations, this study proposes a hierarchical feature selection and optimization framework for endometrial lesions, aiming to enhance the objectivity and robustness of ultrasound-based diagnosis. Firstly, Kernel Principal Component Analysis (KPCA) is employed for nonlinear dimensionality reduction, retaining the top 1000 principal components. Secondly, an ensemble of three filter-based methods—information gain, chi-square test, and symmetrical uncertainty—is integrated to rank and fuse features, followed by thresholding with Maximum Scatter Difference Linear Discriminant Analysis (MSDLDA) for preliminary feature selection. Finally, the Whale Migration Algorithm (WMA) is applied to population-based feature optimization and classifier training under the constraints of a Support Vector Machine (SVM) and a macro-averaged F1 score. Experimental results demonstrate that the proposed closed-loop pipeline of “kernel reduction—filter fusion—threshold pruning—intelligent optimization—robust classification” effectively balances nonlinear structure preservation, feature redundancy control, and model generalization, providing an interpretable, reproducible, and efficient solution for intelligent diagnosis in small- to medium-scale medical imaging datasets.
Accurate assessment of Human Epidermal Growth Factor Receptor 2 (HER2) status in colorectal cancer (CRC) is pivotal for precision therapy, yet the gigapixel resolution of Whole Slide Images (WSIs) presents a significant computational bottleneck for traditional deep learning workflows that rely on exhaustive sliding-window tiling. Addressing this challenge, we propose a novel coarse-to-fine framework that mimics the pathologist's cognitive screening process by integrating swarm intelligence with deep learning. Specifically, we treat the low-magnification WSI as a two-dimensional search space and employ a Particle Swarm Optimization (PSO) algorithm to autonomously navigate and identify diagnostically relevant Regions of Interest (ROIs). The PSO search is guided by a fitness function based on color deconvolution metrics-prioritizing the proportion and intensity of 3,3'-Diaminobenzidine (DAB) staining-thereby effectively filtering out non-informative background and negative tissue without the need for full-slide scanning. In the subsequent stage, these high-value candidate ROIs are extracted at high resolution and analyzed using state-of-the-art deep learning models, such as ResNet or Swin Transformer, to classify HER2 status. Experimental results demonstrate that this swarm intelligence-driven approach reduces computational overhead while achieving favorable patch-level discrimination by focusing analysis on key pathological areas. To clarify how the selected ROI patches behave at the WSI level, we further report per-WSI prediction-count distributions and include an Attention-Based Multiple Instance Learning (ABMIL) baseline. By combining intelligent sampling with deep learning classification, our method provides an interpretable and computationally efficient ROI preselection framework for digital pathology workflows, while broader multi-center validation will be required before clinical deployment.