Bamboo slips were the central medium of the ancient Chinese writing system and serve as key artifacts for understanding the evolution of Chinese characters and civilization. However, due to centuries of complex environmental erosion, these artifacts frequently suffer from severe physical degradation, including ink fading and structural fragmentation. Such irreversible damage severely hinders textual legibility and poses significant challenges to manual interpretation. To address this, we propose the Infrared Structure-Guided Diffusion Network (ICSGDiff), a novel multimodal approach designed for the precise restoration of faded or missing ink on bamboo slips. Our method consists of two key components: the Infrared Character Structure Extraction Module (ICSEM) and the Character Reconstruction Module (CRM). Utilizing the high absorptivity of carbon-based ink in the infrared spectrum, the ICSEM decouples structural features from texture information to generate high-precision binary masks, effectively suppressing background noise such as stains and cracks. Guided by the explicit spatial constraints of these masks, the CRM employs a conditional diffusion model to reconstruct textures in missing regions of visible light images, accurately preserving original stroke patterns and calligraphic styles. Furthermore, to resolve the challenge of broken strokes, we introduce a topology-aware connectivity loss that enforces stroke continuity, ensuring the natural closure of fractured character segments. Experiments on a proprietary bamboo slip dataset demonstrate that ICSGDiff achieves state-of-the-art performance in both quantitative metrics and visual quality. By significantly improving the coherence and readability of degraded text, our work establishes a new paradigm for the digital preservation and intelligent restoration of ancient cultural heritage.

Medical semantic-mask synthesis enhances data augmentation and analysis, yet most GAN-based methods still generate one-to-one images without maintaining spatial consistency across slices. To overcome this, we propose AnatoMaskGAN, a novel framework that embeds interslice spatial features, integrates diverse augmentation strategies, and optimizes deep feature learning for complex medical scans. Specifically, a GNN-based slice-feature fusion module models spatial relationships and aggregates contextual cues from neighboring slices, capturing anatomical structures more comprehensively. A 3D spatial noise-injection strategy fuses weighted spatial features with noise to enrich structural diversity, while a grayscale-texture classifier refines grayscale and texture representation during generation. Experiments on L2R-OASIS and L2R-Abdomen CT show that AnatoMaskGAN achieves 26.50 dB PSNR and 0.9229 SSIM on L2R-OASIS, surpassing the state of the art by 0.43 dB, and reaches 0.8602 SSIM on L2R-Abdomen CT, a 0.48-point improvement. Ablation studies confirm that removing any of the three modules degrades PSNR, SSIM, and LPIPS, verifying the effectiveness of each component in enhancing reconstruction fidelity and perceptual quality. Code is available at https://github.com/noheadwuzonglin/AnatoMaskGAN.

Symbolic scripts carry deep cultural connotations and important historical values. However, due to their unique symbolic structures, linguistic characteristics of weak semantic association and scarce corpus resources, existing neural network machine translation techniques face challenges including insufficient semantic understanding, severe Out-of-Vocabulary issues , poor translation quality, and limited adaptability to semantic noise when handling their translation tasks. To solve this problem, we propose Semantic Enhancement for Weak Semantics Low-Resource Languages Translation method (SEWLT), using the translation task from Naxi Dongba to Chinese as a case study. Experimental results on a self-constructed Naxi Dongba-Chinese small-scale parallel corpus show excellent performance in terms of accuracy, fluency, and semantic fidelity. It not only provides technical support for the digital preservation and research of the Naxi Dongba script, but also provides an important reference for the research of machine translation of similar weak semantic low-resource languages.

Deep subspace clustering based on auto-encoder has received wide attention. However, most subspace clustering based on auto-encoder does not utilize the structural information in the self-expressive coefficient matrix, which limits the clustering performance. In this paper, we propose a double self-expressive subspace clustering algorithm. The key idea of our solution is to view the self-expressive coefficient as a feature representation of the example to get another coefficient matrix. Then, we use the two coefficient matrices to construct the affinity matrix for spectral clustering. We find that it can reduce the subspace-preserving representation error and improve connectivity. To further enhance the clustering performance, we proposed a self-supervised module based on contrastive learning, which can further improve the performance of the trained network. Experiments on several benchmark datasets demonstrate that the proposed algorithm can achieve better clustering than state-of-the-art methods.

This article introduces a text detection model for historical documents images. The handwritten characters in historical documents are always difficult to detect because they contain fuzzy or missing ink, or weathering features and stains; these features will seriously affect the detection accuracy. In order to reduce the influence mentioned above, an effective ATD model is proposed to detect the textbox of characters in historical documents image, and ATD model includes a CNN-based text-box generation network and an NMS-based MSER text-box generation model. As a post-processing method, a text merging algorithm is proposed to achieve higher detection accuracy. The test results on historical document datasets such as Yi, English, Latin, and Italian datasets show that the method in this paper has good accuracy, and it has taken a solid step for the detection of historical documents.

Text detection is the primary task for digitization of ancient books. Different from the common scene text detection tasks (ICDAR, TotalText, etc.), the texts in handwritten ancient documents are more densely distributed and generally small objects; at the same time, the layout structure is also more complex, with problems such as mixed arrangement of pictures and texts and high background noise, all of which pose challenges for detection. According to the characteristics of ancient book images, this paper proposes a new fusion structure based on Feature Pyramid Networks, and takes FCOS as the baseline model to form a new detector (named RFCOS). We enhance the detection capability for dense and small text instances by adding bottom-up fusion paths, cross-layer connections and weighted fusion. Meanwhile, the loss of high-level feature maps during fusion is reduced by new upsampling method and lateral connections. We verified the effectiveness of our RFCOS on the HWAD (Handwritten Ancient Books Dataset), a dataset containing samples in four languages - Yi, Chinese, Tibetan and Tangut, and verify the generalization of RFCOS on another public dataset MTHv2. The results show that RFCOS outperformed most of the existing text detectors in terms of precision, recall and F-measure.

The degree of establishment of cerebrovascular collateral circulation is closely related to the prognosis of patients with acute ischemic stroke, but the evaluation of collateral circulation requires high professional experience of physicians because of the complex structure of the cerebral vessels themselves, and the variety of scoring criteria resulting in poor consistency of results between physicians. Therefore, the use of computer-aided diagnostic techniques to evaluate the establishment of collateral circulation in patients with ischemic stroke is of great clinical importance. In this paper, we proposed a novel method for automatic scoring of collateral circulation via multiphase cranial CTA (computed tomography angiography) to assist physicians in diagnosis. We compared with existing mainstream classification n etworks, our method is able to achieve 90.43% accuracy. Further, the effectiveness of the method was further validated by ablation experiments. However, the multi-phase Cranial CTA collateral circulation scoring algorithm based on a feature fusion network with the hybrid attention mechanism effectively improves the efficiency of prognostic judgment, avoids the limitations of manual extraction of image features in the traditional ways, and plays an auxiliary role in diagnosis for physicians in clinical practice, which is useful for guiding the decision of clinical syndromes in lateral branch circulation stroke.

Acute ischemic stroke (AIS) caused by large vessel occlusion (LVO) has high disability and mortality. However, due to the individual differences of physiological structure and pathological changes between patients, it will be difficult to detect the occluded vessels, so as to delay the treatment timing. Therefore, it is of great significance to a ssist d octors to locate occluded vessels quickly and accurately in clinical practice. In this paper, we present a novel multi-branch hybrid network (MBH-Net) with auxiliary attention guidance to detect occluded vessels. The proposed network consists of three branches for universal representation learning, patient representation learning and classifier learning, respectively. Furthermore, we propose a semantic feature enhancement module to extract more robust semantic information. Particularly, we introduce an auxiliary attention guidance module to guide the attention tendency of MBH-Net, which can make the network give a more reasonable visual interpretation. Extensive experiments show that our MBH-Net can achieve satisfactory accuracy and give a reasonable visual interpretation.

Oracle Bone Inscriptions (OBI) is one of the oldest scripts in the world. The rejoining of Oracle Bone (OB) fragments is of vital importance to the research of ancient scripts and history. Although significant progress has been achieved in the past decades, the rejoining work still heavily relies on domain knowledge and manual work, thus remains a low efficient and time-consuming process Therefore, an automatic and practical algorithm/system for OB rejoining is of great value to the OBI community. To this end, we collect a real-world dataset for rejoining Oracle Bone fragments, namely OB-Rejoin, which consists of 998 OB rubbing images that suffer from low quality image problems, due to intrinsic underground eroding over time and extrinsic imaging conditions in the past. Moreover, a practical Self-Supervised Splicing Network, S3-Net, is proposed to rejoin the OB fragments based on shape similarity of their borderlines. Specifically, we first transform the manually annotated borderline strokes of OB images into times series style shape representations, which are fed as input to a Generative Adversarial Network for augmenting positive pairs of rejoinable OBs for each OB fragment that does not have rejoinable counterparts. A Siamese network is trained on such augmented data in a contrastive learning manner to retrieve the matching OB fragments of an unseen query from an OB fragment gallery. Experiments on the OB-Rejoin benchmark show that our data-driven approach outperforms two recent methods for time-series analysis. In order to demonstrate its practical potential, we deploy the proposed S3-Net method in real tests and ultimately discover dozens of new rejoinings missed by domain experts for decades.

In China, the damage of ancient Yi books are serious. Due to the lack of ancient Yi experts, the repairation of ancient Yi books is progressing very slowly. The artificial intelligence is successful in the field of image and text, so it is feasible for the automatic restoration of ancient books. In this article, a generative adversarial networks with dual discriminator (DDGAN) is designed to restore incomplete characters in the ancient Yi literature. The DDGAN integrates the deep convolution generative adversarial network with an ancient Yi comparison discriminator. Through two training stages, it could iteratively optimizes the ancient Yi character generation networks to obtain the text generator According to the loss of comparison discriminator, DDGAN mode could be optimized. The DDGAN model can generate characters to restore the missing stroke in the ancient Yi. The experiment shows that the proposed method achieves a restoration rate of 77.3% when no more than one third of the characters are missing. This work is effective for the protection of Yi ancient books.

Timely updated road networks are the basis for many real-world applications such as intelligent navigation and traffic management. Existing road updating methods based on remote sensing images learn from historical road databases to update roads. Road extraction models learned from historical images however, are not easily applied to a current image due to spectral differences; and only changed roads need updating. In this paper, an Unchanged-Guided Road Updating (UGRoadUpd) framework is proposed to improve the quality of updated road networks by limiting the road updating range and learning from historical unchanged roads. The UGRoadUpd framework identifies road changes using a novel dual-task dominant-transformer-based neural network for road change detection (DT-RoadCDNet). DT-RoadCDNet executes road segmentation and change detection simultaneously, from bi-temporal remote sensing images. The Dominant-Transformer based Global Context Modeling module in DT-RoadCDNet globally models the contextual spatial structure for improved integrity in roads and road changes. Based on the discovery of road changes, an unchanged-guided road update strategy updates the roads in changed areas by learning from the prior information provided by unchanged roads in a historical road database. Experiments on two newly annotated road change detection and update datasets confirms the effectiveness of our UGRoadUpd framework.

Ancient books that record the history of different periods are precious for human civilization. But the protection of them is facing serious problems such as aging. It is significant to repair the damaged characters in ancient books and restore their original textures. The requirement of the restoration of the damaged character is keeping the stroke shape correct and the font style consistent. In order to solve these problems, this paper proposes a new restoration method based on generative adversarial networks. We use the shape restoration network to complete the stroke shape recovery and the font style recovery. The texture repair network is responsible for reconstructing texture details. In order to improve the accuracy of the generator in the shape restoration network, we use the adversarial feature loss (AFL), which can update the generator and discriminator synchronously to replace the traditional perceptual loss. Meanwhile, the font style loss is proposed to maintain the stylistic consistency for the whole character. Our model is evaluated on the datasets Yi and Qing, and shows that it outperforms current state-of-the-art techniques quantitatively and qualitatively. In particular, the Structural Similarity has increased by 8.0% and 6.7% respectively on the two datasets.

The automatic identification of hematoma expansion signs is very important to the diagnosis and treatment of intracerebral hemorrhage. However, the brain computed tomography samples is uneven distribution, low variation in imaging performance between signs, developing such a solution is challenging.In this article, a novel deep learning network is presented to recognize hematoma expansion signs automatically, which provides a new method to assist in the diagnosis of hematoma growth. First, we propose the fusion branching network (FBN) for extracting and fusing the high-dimensional features of each category dynamically. Second, this paper designs a Class Learning Difficulty (CLD) loss function to dynamically adjust the class weights based on the FBN learning situation to help the model avoid learning bias and improve the performance in signs recognition. In addition, a visualization component is provided to improve the transparency of the model. In the experiment, we retrospectively collected DICOM images from the First Affiliated Hospital of C hongqing Medical University, in order to form a recognition process consistent with clinical work, the data is divided into three categories: Blend Sign (BS), Black Hole Sign (BHS), ICH without BS and BHS (ICHWBB) for sign recognition. The experimental results demonstrate that our method performed the best in terms of comprehensive evaluation indexes compared with other methods, and the recognition sensitivity reached 0.9404, 0.8056, 0.7586 for ICHWBB, BHS, and BS, respectively, which was significantly improved compared with other methods in the same environment.

In this letter, a novel neural network (CDMI-Net) that combines change detection and multiple instance learning (MIL) is proposed for landslide mapping. After obtaining a score map of landslides provided by the network, the final binary map is generated by fast postprocessing. The benefits of the proposed method are threefold. First, using the MIL framework, the network is trained only by the scene-level samples and it reduces the need for pixel-level samples. Second, a change-detection network architecture using a two-stream U-Net with shared weights is designed to learn the deep features of the landslide from the two-period aerial images, reducing the false-positive results. Third, integrating a gated attention-based pooling layer and a fast level-set evolution algorithm can finally produce the pixel-level results. Experimental results show that the proposed CDMI-Net achieves comparable and even better performance on the testing image pairs than all other methods and has great potential for the landslide mapping application.

The rejoining of oracle bone rubbings is a fundamental topic for oracle research. However, it is a tough task to reassemble severely broken oracle bone rubbings because of detail loss in manual labeling, the great time consumption of rejoining, and the low accuracy of results. To overcome the challenges, we introduce a novel CFDA&CAP algorithm that consists of the Curve Fitting Degree Analysis (CFDA) algorithm and the Correlation Analysis of Pearson (CAP) algorithm. First, the orthogonalization system is constructed to extract local features based on the curve features analysis. Second, the global feature descriptor is depicted by using coordinate points sequences. Third, we screen candidate curves based on the features as well as the CFDA algorithm, so the search range of the candidates is narrowed down. Finally, image recommendation libraries for target curves are generated by adopting the CAP algorithm, and the rank for each target matching curve generates simultaneously for result evaluation. With experiments, the proposed method shows a good effect in rejoining oracle bone rubbings automatically: (1) it improves the average accuracy rate of curve matching up to 84%, and (2) for a low-resource task, the accuracy of our method has 25% higher accuracy than that of other methods.

It is a technological challenge to recognize landslides from remotely sensed (RS) images automatically and at high speeds, which is fundamentally important for preventing and controlling natural landslide hazards. Many methods have been developed, but there remains room for improvement for stable, higher accuracy, and high-speed landslide recognition for large areas with complex land cover. In this article, a novel integrated approach combining a deep convolutional neural network (CNN) and change detection is proposed for landslide recognition from RS images. Logically, it comprises the following four parts. First, a CNN for landslide recognition is built based on training data sets from RS images with historical landslides. Second, the object-oriented change detection CNN (CDCNN) with a fully connected conditional random field (CRF) is implemented based on the trained CNN. Third, the preliminary CDCNN is optimized by the proposed postprocessing methods. Finally, the results are further enhanced by a set of information extraction methods, including trail extraction, source point extraction, and attribute extraction. Furthermore, in the implementation of the proposed approach, image block processing and parallel processing strategies are adopted. As a result, the speed has been improved significantly, which is extremely important for RS images covering large areas. The effectiveness of the proposed approach has been examined using two landslide-prone sites, Lantau Island and Sharp Peak, Hong Kong, with a total area of more than 70 km2. Besides its high speed, the proposed approach has an accuracy exceeding 80%, and the experiments demonstrate its high practicability.

The digital PCR instrument is a digital instrument for amplifying specific DNA fragments. The problem studied in this paper is the autofocus problem of its electronic imaging device. Based on the analysis of existing SOM neural network autofocus scheme, we propose an improved scheme-BP neural network for autofocus. It directly takes the SOM input and the actual focus position as the input and output of the BP neural network, which eliminates the process of prior classification and then corresponding to the focus matrix in the original SOM scheme, saving time. The experimental results show that the traditional autofocus method has good focusing effect, but the speed is slow, and the universality of the BP neural network autofocus scheme is not good enough, but within a good accuracy range, the speed is faster. Compared to traditional focusing methods, the autofocus scheme designed in this paper successfully achieves faster focusing speed for biochips.

This paper proposes a novel scheduling technique at the layer of block I/O in the storage stack, to reduce the time required for completing I/O requests. Considering the fact of I/O requests issued by different applications may target at different ranges of data blocks on the disk (e.g. sectors in a hard disk drive), we present an adaptive splitting mechanism to preferably group relevant block requests, by referring their offsets (i.e. block numbers). Then, it performs the optimization task of merging consecutive block requests within each group, since only the block requests in the same group are most likely to be associated with a specific application. Finally, the well-scheduled block requests will be forwarded to the disk storage, to be eventually fulfilled. Through a series of experiments on several block traces of real-world multimedia applications, we show that compared to the existing block I/O scheduling approaches, the proposed scheme cuts down the I/O response time by more than 10.4%.