The resolution of feature maps is a critical factor for accurate medical image segmentation. Most of the existing Transformer-based networks for medical image segmentation adopt a U-Net-like architecture, which contains an encoder that converts the high-resolution input image into low-resolution feature maps using a sequence of Transformer blocks and a decoder that gradually generates high-resolution representations from low-resolution feature maps. However, the procedure of recovering high-resolution representations from low-resolution representations may harm the spatial precision of the generated segmentation masks. Unlike previous studies, in this study, we utilized the high-resolution network (HRNet) design style by replacing the convolutional layers with Transformer blocks, continuously exchanging feature map information with different resolutions generated by the Transformer blocks. The proposed Transformer-based network is named the high-resolution Swin Transformer network (HRSTNet). Extensive experiments demonstrated that the HRSTNet can achieve performance comparable with that of the state-of-the-art Transformer-based U-Net-like architecture on the 2021 Brain Tumor Segmentation dataset, the Medical Segmentation Decathlon’s liver dataset, and the BTCV multi-organ segmentation dataset.

Compared with static optical imaging, dynamic optical imaging technology can obtain quantitative pharmacokinetics information, such as the probe metabolism curve, removal rate, and binding potential of the receptor. Accurate segmentation of the region of interest (ROI) is an important step in dynamic optical imaging. Generally, the ROI is manually labeled by researchers based on experience. This will lead to two unavoidable problems. First, manual segmentation of the ROI is very time consuming, especially when there are many sequential dynamic optical images. Second, manual segmentation cannot ensure accuracy when the optical signal gradually decays to a point at which it is difficult to distinguish by using the naked eyes. These problems will inevitably lead to inaccuracy of quantitative results of dynamic optical imaging. Here, we presented a machine learning-based automatic segmentation method to avoid these time-consuming and inaccuracy problems caused by manual segmentation. The K-means clustering algorithm and fuzzy c-means clustering algorithm were implemented to separate the ROI from the background of sequential dynamic optical images. Automatic selection of clustering results was completed by mathematical methods. The accuracy and feasibility of machine learning-based methods were verified by comparing their results with the manual segmentation results. The preliminary results demonstrated that the machine learning-based automatic segmentation has coherent performance with the manual one.

With the emergence of cloud-native computing, serverless computing has become a popular way to deploy intensive applications due to its scalability and flexibility, and it has been increasingly applied in the field of big data processing on service platforms. Currently, the development momentum of cloud-native computing is strong, and serverless computing has become more attractive to the growing number of Internet services. However, how to more effectively address the issues of container resource usage and service startup time for serverless computing remains a huge challenge when exploring its potential. Our research is based on the complete life cycle of serverless functions and improves the performance of serverless computing by changing the original method of exchanging space for time or time for space. We focus on how to shorten the cold-start time of serverless computing while maximizing the usage of container resources. The research innovation is the dynamic control of functions and container pools, which mainly includes three aspects. First, we create a container pool with the classification identification based on the usage rate of functions. Then, we use namespace technology to achieve container resource reuse in the security isolated state. Next, we adaptively match the correspondence between functions and reusable container resources through system resource monitoring. Finally, the test results prove that converting the remaining space resources of the container into a prewarm container for new functions can effectively reduce the resource waste caused by idle function-containers, and container resource reuse can further shorten the cold-start time while ensuring the safety and isolation of functions. Compared to other open-source serverless platforms, our solution can reduce the cold-start time of general function calls to less than 20 ms and improve the ability to alleviate cold starts by 90% without enabling container prewarming.

Traditional deep learning models for fruit and vegetable classification are usually implemented via training on an unchanged dataset. However, changing fruit and vegetable categories is a very common occurrence in the context of real agricultural sales. When dealing with changes related to variety, deep learning models need to be retrained on the entire updated dataset. The retraining process is time-consuming and inefficient, and it may even cause the ‘catastrophic forgetting’ problem. In response to this challenge, the Adversarial Domain Adaptation Class Incremental Learning (ADA-CIL) method is introduced. This approach employs adversarial domain adaptation techniques combined with core-set selection strategies to effectively extract and integrate cross-domain features. We utilize the ResNet34 architecture as the backbone for feature extraction due to its deep residual learning framework, which is robust in handling the complexities of large and varied image datasets. It achieves a dynamic balance in learning between new and existing categories, significantly enhancing the model’s generalization capabilities and information retention efficiency. The FruVeg dataset, composed of three sub-datasets, includes over 120,000 color images, covering more than 100 different categories of fruits and vegetables collected from various domains and backgrounds. The experimental results on the FruVeg dataset show that the ADA-CIL method achieves an average accuracy of 96.30%, a forgetting rate of 2.96%, a cumulative accuracy of 96.26%, and a current accuracy of 98.60%. The ADA-CIL method improves the average accuracy by 1.65% and 1.82% compared to iCaRL and BiC, respectively, and it reduces the forgetting rate by 2.69% and 2.76%. These performance metrics demonstrate the ADA-CIL method’s impressive ability to handle incremental category and domain changes, highlighting its capability to effectively maintain the intra-class stability and exhibit exceptional adaptability in dynamic learning environments.

Extracting objects of interest from remote sensing imagery is an essential part in various practical applications. The objects that people pay attention to in the remote sensing scene mainly include buildings, roads, vehicles, etc. In this article, extracting the aforementioned objects are collectively referred to as the target extraction task. Arising from object scale variation, appearance similarity between adjacent patches, diversity of imaging orientation, and complexity of background, it is difficult to extract complete objects from cluttered backgrounds. Deep neural network has made great achievement in dense prediction for target extraction. However, most of the previous works are still faced with a formidable challenge in discriminative context feature representation to extract targets of various categories and correctly classify pixels around the boundary. In this article, we propose a target extraction neural network, named discriminative context-aware network, to focus on discriminative high-level context features and preserve spatial information. First, a discriminative context-aware feature module is designed to generate the feature maps in the top layer, which not only captures the rich image context information but also aggregates the contrasted local information at multiple scales. Second, a refine decoder module is adopted to preserve spatial information from low-level layers and enhance the feature representation, leading to precise segmentation results. We conducted extensive experiments on building and road extraction benchmarks, including WHU building dataset and Massachusetts road dataset, together with a self-constructed dataset for vehicle extraction in SAR images. Our method achieves state-of-the-art results with fewer parameters and faster inference.