Medical image segmentation is increasingly reliant on deep learning techniques, yet the promising performance often come with high annotation costs. This paper introduces Weak-Mamba-UNet, an innovative weakly-supervised learning (WSL) framework that leverages the capabilities of Convolutional Neural Network (CNN), Vision Transformer (ViT), and the cutting-edge Visual Mamba (VMamba) architecture for medical image segmentation, especially when dealing with scribble-based annotations. The proposed WSL strategy incorporates three distinct architecture but same symmetrical encoder-decoder networks: a CNN-based U-Net for detailed local feature extraction, a Swin Transformer-based Swin-UNet for comprehensive global context understanding, and a VMamba-based Mamba-UNet for efficient long-range dependency modeling. The key concept of this framework is a collaborative and cross-supervisory mechanism that employs pseudo labels to facilitate iterative learning and refinement across the networks. The effectiveness of Weak-Mamba-UNet is validated on two publicly available datasets with processed scribble annotations, where it surpasses the performance of a similar WSL framework utilizing only U-Net or Swin-UNet, as well as other baseline methods. This paper highlights the potential of Mamba for medical image segmentation in scenarios with sparse or imprecise annotations. The source code, dataset, and all baseline methods are made publicly accessible https://github.com/ziyangwang007/Mamba-UNet.
Advanced manufacturing is a new trend for sustainable industrial development, and digital twin is a new technology that has attracted attention. Blast furnace smelting is an effective method in the manufacturing of iron and steel. Comprehensive and dependable surveillance of the blast furnace smelting process is essential for ensuring the smooth operation and improving of iron and steel output quality. The current technology makes it difficult to monitor the entire process of blast furnace ironmaking. Based on Unity 3D, this study presents a digital-twin virtual reality simulation system of blast furnace ironmaking. First, shading modeling creates a three-dimensional dynamic geometric model in different ironmaking system scenarios. Then, we script the animation and call particle system according to the motion mode of distinct geometric objects to give the dynamic effect of geometric objects. Shaders are the focus of the design and contributions. In addition, shader optimization technology can reduce hardware resource consumption and increase system fluency. Vertex shaders are used for all types of coordinate space transformation and vertex output; fragment shaders are used for texture sampling, light model calculation, normal calculation, noise superposition, and color output. The shader rendering technique allows for more realistic lighting effects. The presented dynamic digital twin system implements more realistic lighting analyzed in the ironmaking process. Virtual interaction logic’s design and deployment process is based on HTC VIVE hardware and VRTK toolkit. In the actual simulation process, the typical animation frame rate is stable at about 75 FPS (frames per second). The simulation system runs smoothly and a cutting-edge and state-of-the-art method for observing the blast furnace ironmaking process is suggested.
In the aluminum reduction process, the flame hole is an influential index of the whole production situation, which reflects the distribution of the physical field of the reduction cell, the current efficiency and the lifespan of the aluminum reduction cell. Therefore, flame hole situation detection and identification are critical and significant in the whole process of aluminum electrolysis. However, in the practical industrial production, flame hole identification result is coming from the experimental operation workers, with the loss of workers and different experimental levels, the real-time measurement of the flame hole index is still a challenge beyond solution. This article develops a flame hole image classification method based on wavelet extreme learning machine. First, a deep feature set is extracted from the original images with a convolutional neural network. Then, a classification model based on an extreme learning machine with wavelet activation function is developed. Finally, the proposed convolutional neural network–weighted extreme learning machine model is applied to superheat degree real-time detection in the industrial electrolysis cell. The proposed method is evaluated on aluminum production which outperforms existing other superheat degree methods in accuracy and robustness.