The scarcity of parallel corpora for Mongolian and Chinese constrains the performance of Mongolian-Chinese neural machine translation (NMT), particularly manifesting in inadequate accuracy in translating specialized terminology. To address this limitation, this study adopts a lexically constrained augmentation strategy that constructs pseudo-source sentences by appending Chinese constraint words to Mongolian source texts, while enforcing the inclusion of these constraints in the output to improve translation accuracy. However, this approach presents two inherent drawbacks: processing pseudo-sentences with a single encoder tends to induce semantic interference, while the introduced constraint words may exacerbate alignment errors during decoding. To overcome these limitations, this paper propose a Constraint-Augmented Mongolian-Chinese NMT method (CANMT) based on dynamic feedback alignment. The method employs a dual-encoder architecture to isolate bilingual representations, coupled with a dynamic feedback alignment module that progressively reduces alignment errors through iterative reffnement, thereby enhancing overall translation performance.

Logit Knowledge Distillation has gained substantial research interest in recent years due to its simplicity and lack of requirement for intermediate feature alignment; however, it suffers from limited interpretability in its decision-making process. To address this, we propose implicit Clustering Distillation (iCD): a simple and effective method that mines and transfers interpretable structural knowledge from logits, without requiring ground-truth labels or feature-space alignment. iCD leverages Gram matrices over decoupled local logit representations to enable student models to learn latent semantic structural patterns. Extensive experiments on benchmark datasets demonstrate the effectiveness of iCD across diverse teacher-student architectures, with particularly strong performance in fine-grained classification tasks---achieving a peak improvement of +5.08% over the baseline.

Digitizing historical Mongolian newspapers is challenging due to severe degradation and unique script properties. Elongated words and artifacts, like ink bleeding cause geometric distortion. These issues limit traditional OCR performance. To address these issues, we propose MGL-LMM, a Generative OCR Paradigm specifically designed for degraded agglutinative scripts. Our framework integrates a topology-preserving visual normalization strategy. This maintains structural integrity under anisotropic distortion. To stabilize training, we employ a decoupled curriculum learning scheme that transitions from word-level glyph morphologies to line-level contextualization. By projecting high-fidelity visual features directly into the semantic space of a 1.5B-scale Large Language Model, MGL-LMM performs end-to-end generative reasoning. This allows the model to leverage contextual language priors to effectively reduce visual ambiguities and provide semantic restoration for obscured characters. Extensive experiments demonstrate that our method achieves state-of-the-art performance, reducing the Character Error Rate (CER) from 27% to 11%. This significant improvement demonstrates that generative decoding with language priors is fundamentally more effective than purely visual recognition under severe degradation.

The classification of highly complex and detail-rich images, which is essential for botanical research and environmental monitoring, presents a significant challenge in electron microscope pollen analysis due to the need to enhance both accuracy and computational efficiency. This study introduces the SNT model, In order to achieve this, it incorporates Large-view Spatial Bilateral Attention (LSA), which is adept at capturing both local details and global contextual information. A novel lightweight pruning technique is employed to retain essential weight information while significantly reducing the model’s size, thereby enhancing its utility and efficiency. The results of our experimental study demonstrate that the SNT model exhibits a significant enhancement in accuracy compared to the advanced CNNs and Transformer model.

Anomaly detection plays a crucial role in quality control within the manufacturing industry. Traditional computer vision approaches typically adopt reconstruction-based strategies, where the distribution of normal features is modeled to suppress the reconstruction of anomalous regions. Although these methods generally require fewer data, they often struggle to simultaneously improve reconstruction quality and detection performance. Experimental results indicate that high-level semantic context features exhibit strong capabilities in identifying anomalies, while low-level features contribute to the representation of global characteristics. Based on this, the paper uses discrete wavelet transform (DWT) to decompose images into low- and high-frequency components, extracting both global and local features. The input image is encoded in parallel using convolutional encoders and fused with the base features extracted through DWT. To further enhance detection performance, we propose a customized feature fusion module inspired by Feature Pyramid Networks (FPN), which integrates spatial and channel attention mechanisms to strengthen the model's ability to capture inter-feature relationships. Finally, extensive comparative and ablation experiments conducted on public benchmark datasets demonstrate the effectiveness of the proposed method.

Detecting electron microscope images of cashmere and wool fibers is challenging; traditional manual methods struggle to distinguish between them. Cashmere and wool fibers are extremely similar in scale, fiber length, and fineness, coupled with the scarcity of relevant resources caused by the exorbitant cost of obtaining electron microscope images. These factors are intertwined, making it also difficult for the accuracy of deep learning detection methods to reach an ideal level. In response to the preceding issue, this paper presents a detector capable of learning discriminative features between fibers from electron microscope images and effectively completing the detection task with limited data. Initially, to address the challenge of capturing the complex features of cashmere and wool, this paper introduces Wool-Linear Deformable Convolution (W-LDConv) derived from LDConv. Furthermore, to overcome issues such as the inability of deformable large-kernel attention to adapt to fiber distribution and the excessive number of parameters, W-LDConv is used to take over the role of deformable convolution, forming the Linear Deformable Feature Capture Module (LDFC). Ultimately, the LDFC module is integrated with the efficient and stable YOLO11 model to propose the LDFC-YOLO object detector. Compared with the baseline model, the precision, recall, and [email protected] have improved by 12.9%, 18.6%, and 13.9%, respectively.

Neural Machine Translation (NMT) has achieved significant progress for high-resource language pairs but still faces challenges with low-resource pairs like Mongolian-Chinese. Mongolian presents unique grammatical and semantic modeling difficulties as an agglutinative language with rich morphology and SOV word order (Subject-Object-Verb). Taking Convolutional Neural Networks (CNNs) and Transformers as examples: CNNs struggle to capture long-range dependencies, while Transformers, despite their global modeling capabilities, impose high computational costs. Existing Mongolian-Chinese NMT methods often prioritize translation quality but overlook computational efficiency, limiting their applicability on resource-constrained devices. This paper proposes a lightweight model, Dynamic Convolution with Long-Range Attention (DCLA), which balances translation quality and efficiency. DCLA uses a recognizer module to analyze Mongolian-specific features, such as sentence length, structural complexity, and low-frequency word distribution. It effectively addresses syntactic challenges caused by SOV-SVO word order differences and semantic difficulties posed by polysemous and low-frequency words. DCLA adopts two strategies based on sample complexity: applying convolution for simpler samples to reduce computational cost and multi-head attention for complex samples to enhance modeling. Experiments demonstrate that DCLA outperforms the state-of-the-art Transformer model in Mongolian-Chinese translation, with a 2.7-point BLEU improvement, a 21% reduction in parameters, and a 40% decrease in computational cost.

Mongolian speech synthesis is a technology that converts Mongolian text into Mongolian speech. In order to improve the emotional expressiveness of synthesized speech, this article first proposed a lightweight Mongolian phoneme pre-training model WFST-MnG2P based on weighted finite state transition machine. Secondly, as a representative low-resource language, Mongolian currently has no open source emotional speech corpus. For this reason, a Mongolian emotional speech corpus containing seven discrete emotions was constructed, totaling about 2.25 hours. Finally, since the non-autoregressive acoustic model can reduce word skipping, word missing, repeated pronunciation, and so on, and speed up the speech synthesis speed, this article proposes a Mongolian emotional speech synthesis model based on conditional generative adversarial network and improved FastSpeech2. Experimental results show that the average MOS score of emotional speech on the self-built Mongolian emotional speech corpus is 3.69, and the model can synthesize Mongolian emotional speech with rich multi-dimensional emotions and more robustness.

The characteristics of neural machine translation require training and updating through a large number of corpus for hundreds of millions of parameters. In this situation, Mongolian Neural Machine Translation(MNMT) needs a variety of targeted training techniques and additional strategies to alleviate the problems caused by resource scarcity. These problems run through the whole translation process. taking the key steps of model training as a clue, This paper conducted detailed experiments and analysis on the main training content, and summarizes the relevant research and key issues according to the mainstream training processes such as ‘corpus processing→word embedding training→parameter pre-training→end-to-end model training→translation key problem analysis’. On this basis, this paper is committed to some long-standing stubborn problems to give a review of the treatment methods and training suggestions, and to provide some references for other researchers.

Conventional Neural Machine Translation (NMT) relies on previous tokens and the hidden state of the target for the inference of the target tokens in the decoding phase, and this left-to-right decoding approach loses the context of the target sequence from right-to-left. In addition, the vanilla attention mechanism lacks interactivity in the bilingual training phase, where the computation of the attention weights is independent at each step, which leads the decoder to disregard whether or not the current token has already been translated. In this paper, we proposed a novel interactive synchronized bi-directional inference method that uses past and future contexts to synchronously predict target sequences and incorporates Neural Turing Machine (NTM) ideas to detect historical attentional information, which relies on a read − write mechanism to update the source hidden state. We evaluate the proposed model in the WMT14 German-English translation task and the LDC Chinese-English translation task, and the experimental results show that our method improves 2.18 and 4.07 BLEU scores over the Transformer, respectively, which fully demonstrates the effectiveness of the proposed method.