With the advancement of educational informatization, vast amounts of Chinese text are generated across online platforms and digital textbooks. Effectively classifying such text is essential for intelligent education systems. This study conducts a systematic comparative evaluation of three Transformer-based models—TinyBERT-4L, BERT-base-Chinese, and RoBERTa-wwm-ext—for Chinese educational text classification. Using a balanced four-category subset of the THUCNews corpus (Education, Technology, Finance, and Stock), the research investigates the trade-off between classification effectiveness and computational efficiency under a unified experimental framework. The experimental results show that RoBERTa-wwm-ext achieves the highest effectiveness (93.12% Accuracy, 93.08% weighted F1), validating the benefits of whole-word masking and extended pre-training. BERT-base-Chinese maintains a balanced performance (91.74% Accuracy, 91.66% F1) with moderate computational demand. These findings reveal a clear symmetry–asymmetry dynamic: structural symmetry arises from the shared Transformer encoder and identical fine-tuning setup, while asymmetry emerges from differences in model scale and pre-training strategy. This interplay leads to distinct accuracy–latency trade-offs, providing practical guidance for deploying pre-trained language models in resource-constrained intelligent education systems.

Large language models and retrieval-augmented generation (RAG) are widely applied in knowledge question-answering tasks. However, in knowledge-intensive domains such as agriculture, hallucination and insufficient retrieval accuracy remain challenging. To address these issues, we propose Sem-RAG, a corn planting knowledge question-answering algorithm based on fine-grained semantic retrieval enhancement. Unlike standard NaiveRAG, which retrieves only fixed-length text chunks, and GraphRAG, which relies solely on graph node connections, Sem-RAG introduces a dual-store retrieval mechanism. It constructs both a surface semantic store (chunk-level embeddings) and a fine-grained semantic store derived from Leiden-based community summaries. These community summaries do not merely shorten contexts; instead, they provide thematic-level semantic aggregation across document chunks, thereby enhancing semantic coverage and reducing noise. During retrieval, user queries are matched against the surface store to locate relevant chunks and simultaneously linked to corresponding thematic summaries in the fine-grained store, ensuring that both local details and higher-level associations are leveraged. We evaluated Sem-RAG on the corn knowledge question-answering dataset CornData. The algorithm achieved Answer-C, Answer-R, and CR scores of 94.6%, 84.6%, and 70.4%, respectively, which were 2.6%, 1.7%, and 1.6% higher than those of traditional NaiveRAG. These results demonstrate that Sem-RAG materially improves the quality and reliability of agricultural knowledge Q&A by combining dual-store retrieval with community-level semantic aggregation.

Small-target detection in Unmanned Aerial Vehicle (UAV) aerial images remains a significant and unresolved challenge in aerial image analysis, hampered by low target resolution, dense object clustering, and complex, cluttered backgrounds. In order to cope with these problems, we present AeroLight, a novel and efficient detection architecture that achieves high-fidelity performance in resource-constrained environments. AeroLight is built upon three key innovations. First, we have optimized the feature pyramid at the architectural level by integrating a high-resolution head specifically designed for minute object detection. This design enhances sensitivity to fine-grained spatial details while streamlining redundant and computationally expensive network layers. Second, a Dynamic Feature Fusion (DFF) module is proposed to adaptively recalibrate and merge multi-scale feature maps, mitigating information loss during integration and strengthening object representation across diverse scales. Finally, we enhance the localization precision of irregular-shaped objects by refining bounding box regression using a Shape-IoU loss function. AeroLight is shown to improve mAP50 and mAP50-95 by 7.5% and 3.3%, respectively, on the VisDrone2019 dataset, while reducing the parameter count by 28.8% when compared with the baseline model. Further validation on the RSOD dataset and Huaxing Farm Drone dataset confirms its superior performance and generalization capabilities. AeroLight provides a powerful and efficient solution for real-world UAV applications, setting a new standard for lightweight, high-precision object recognition in aerial imaging scenarios.

As one of the world’s leading cotton-producing countries, China frequently experiences severe yield reductions due to crop diseases and pest infestations, with losses often exceeding 20%. Although computer vision models can identify diseased plants, they currently fail to connect visual symptoms to the diagnostic reasoning process used by agronomists. This leads to text descriptions that ignore the biological causes of the damage. To fix this, we built Multimodal Knowledge Graph-Enhanced Cross Vision Transformer-18-Dagger-408 and Text-to-Text Transfer Transformer for Cotton Disease and Pest Image Captioning (MKG-CottonCapT6), a model that uses a local knowledge database to generate professional diagnostic reports from field images. The technical core consists of a Multimodal Knowledge Graph (MMKG) containing 14 types of entities (such as Pathogens and Control Agents) and 12 types of relations. We use a Cross Vision-Transformer-18-Dagger-408 (CrossViT) encoder to capture both the overall leaf shape and microscopic details of pests. Through a Visual Entity Grounding (VEG) module, the model maps visual features directly to specific triplets in the graph. These triplets are then turned into text sequences and fused with image data in a Text-to-Text-Transfer-Transformer (T5) decoder. To train the model, we collected a dataset of cotton images paired with expert descriptions of lesions, colors, and affected plant parts. Tests show that MKG-CottonCapT6 performs better than standard models, reaching an Information-based Metric for Image Captioning (InfoMetIC) score of 72.6%. Results prove that by using a specific alignment loss (${\mathcal{L}}_{align}$), the model generates reports that correctly name the disease stage and recommend specific chemicals, such as Carbendazim or Triadimefon. This framework provides a practical tool for farmers to record and treat cotton diseases with high precision.

Distant supervised relation extraction (DSRE) is widely used to extract novel relational facts from plain text, so as to improve the knowledge graph. However, distant supervision inevitably suffers from the noisy labeling problem that will severely damage the performance of relation extraction. Currently, most DSRE methods are mainly focused on reducing the weights of noisy sentences, ignoring the bag-level noise where all sentences in a bag are wrongly labeled. In this paper, we present a novel noise detection-based relation extraction approach (NDRE) to automatically detect noisy labels with entity information and dynamically correct them, which can alleviate both instance-level and bag-level noisy problems. By this means, we can extend the dataset from the Web tables without introducing more noise. In this approach, to embed the semantics of sentences from corpus and web tables, we firstly propose a powerful sentence coder that employs an internal multi-head self-attention mechanism between the piecewise max-pooling convolutional neural network. Second, we adopt a noise detection strategy, which is expected to dynamically detect and correct the original noisy label according to the similarity between sentence representation and entity-aware embeddings. Then, we aggregate the information from corpus and web tables to make the final relation prediction. Experimental results on a public benchmark dataset demonstrate that our proposed approach achieves significant improvements over the state-of-the-art baselines and can effectively reduce the noisy labeling problem.

The identification of cotton diseases and pests is crucial for maintaining cotton yield and quality. However, conventional manual methods are inefficient and prone to high error rates, limiting their practicality in real-world agricultural scenarios. Furthermore, Convolutional Neural Network–Long Short-Term Memory (CNN-LSTM) models are insufficient in generating fine-grained and semantically rich image captions, particularly for complex disease and pest features. To overcome these challenges, we introduce CottonCapT6, a novel multi-task image captioning framework based on the Cross Vision Transformer (CrossViT-18-Dagger-408) and Text-to-Text Transfer Transformer (T5). We also construct a new dataset containing annotated images of seven common cotton diseases and pests to support this work. Experimental results demonstrate that CottonCapT6 achieves a Consensus-based Image Captioning Evaluation (CIDEr) score of 197.2% on the captioning task, demonstrating outstanding performance. Notably, the framework excels in providing more descriptive, coherent, and contextually accurate captions. This approach has strong potential to be deployed in cotton farms in the future, helping pest control personnel and farmers make precise judgments on cotton diseases and pests. However, its generalizability to other crops and environmental conditions remains an area for future exploration.