The tear strength of textiles is a crucial characteristic of product quality. However, during the laboratory testing of this indicator, factors such as equipment operation, human intervention, and test environment can significantly influence the results. Currently, there is a lack of traceable records for the influencing factors during the testing process, and effective classification of testing activities is not achieved. Therefore, this study proposes a state-awareness and classification approach for fabric tear performance testing based on multi-source data. A systematic design is employed for fabric tear performance testing activities, which can real-time monitor electrical parameters, operational environment, and operator behavior. The data are collected, preprocessed, and a Decision Tree Support Vector Machine (DTSVM) is utilized for classifying various working states, and introducing ten-fold cross-validation to enhance the performance of the classifier, forming a comprehensive awareness of the testing activities. Experimental results demonstrate that the system effectively perceives fabric tear performance testing processes, exhibiting high accuracy in the classification of different fabric testing states, surpassing 98.73%. The widespread application of this system contributes to continuous improvement in the workflow and traceability of fabric tear performance testing processes.

The capability parameters of third-party testing institutions not only serve as a critical reflection of their technical and quality management capabilities but also form the key basis for categorizing their testing abilities. However, current Chinese phrase-based descriptions of these capability parameters are influenced by diverse expression styles and varying internal standards, making it difficult to establish consistent criteria for classifying testing capabilities. This inconsistency presents notable difficulties for clients and regulatory bodies. Therefore, leveraging clustering techniques to uncover the intrinsic relationships and latent information between testing capabilities and their corresponding parameters is one of the crucial approaches to achieving scientific and reasonable classification of testing capabilities. Traditional text feature extraction methods suffer from several limitations, including sparse features, high-dimensional features, and lack of semantic information. These shortcomings complicate the classification and analysis of testing capability descriptions. To address this issue, this study focuses on the “products and testing objects” within the capability parameters of Chinese testing institutions as the research subject and proposes a joint model based on BERT and semi-supervised K-Means clustering. This model employs BERT to extract textual features from Chinese descriptive phrases and combines them with a small number of labeled samples for semi-supervised K-Means clustering analysis. The clustering results are then used to train a multi-output Softmax classifier, thereby enabling the classification of testing capabilities for third-party institutions. Experimental results demonstrate that the proposed method outperforms traditional methods such as TF-IDF and one-hot encoding when applied to the Chinese description datasets of testing institutions. Specifically, it exhibits advantages in reducing the dimensionality of textual features and enhancing clustering performance. When the proportion of labeled samples accounts for 10% of the total sample size, the method achieves optimal clustering results, with an average classifier accuracy of 89.8%.

The degree of pilling and fuzzing in textile fabrics is a crucial indicator of textile product quality. Current evaluation methods predominantly rely on subjective judgments, leading to issues such as rating errors and inefficiency. To achieve objective assessment of pilling and fuzzing grades, this study proposes a Hybrid Feature-Based Machine Vision Method for Objective Evaluation of Textile Pilling and Fuzzing. The method incorporates a Hybrid Feature-based Depthwise Separable Attention Network for Objective Evaluation of Textile Pilling and Fuzzing (HDAN-PF), which effectively extracts and fuses frequency and Space domain features. A Channel Attention mechanism enhances the model’s ability to capture subtle features, while Depthwise Separable Convolutions reduce computational complexity, improving evaluation speed while maintaining high accuracy.The model size is approximately 327.37 MB with a total parameter count of 135,115,512. Experimental results demonstrate that the proposed method achieves a classification accuracy of 96.26% on diverse fabric images, showcasing robust generalization and practical utility.By leveraging this machine vision approach, the proposed method offers a transformative solution for achieving objective, consistent, and efficient assessments of pilling and fuzzing grades, advancing textile quality evaluation practices.

Fabric pilling performance is one of the key indicators for evaluating textile quality, but there is limited research on the effectiveness of pilling detection traceability and non-intrusive monitoring of detection equipment operating status. In this paper, a multi-source data-driven method for classifying the working status of fabric pilling performance detection is proposed. This study constructs a real-time non-intrusive monitoring system for fabric pilling detection in a laboratory environment, collecting multiple types of data such as electrical parameters of pilling detection equipment, personnel behavior, and equipment noise. GoogLeNet convolutional neural network is used to recognize high-dimensional audio data and achieve feature dimension reduction. By constructing a multi-classification algorithm based on Decision Tree-Support Vector Machine (DT-SVM) for the pilling detection process, a minimum accuracy of 95.62% is achieved in practical operation. This system not only perceives the relevant influencing factors of detection activities without interfering with normal detection activities but also effectively distinguishes various detection working states, providing new ideas for the effectiveness traceability of pilling detection activities.

With the extensive application of wood-based panels in the home decor industry, determining formaldehyde release form wood-based panels (FRWBP) is crucial for ensuring environmental air quality and human health. This article addresses the issues of insufficient monitoring of the testing stages for FRWBP and the difficulty in tracing the operation records of operator during determining FRWBP by the chamber method. It proposes a system for monitoring and analyzing the testing stages of FRWBP based on electrical monitoring. The system employs non-invasive sensing technology that does not interfere with normal testing process to collect real-time electrical usage data from the devices. The data is uploaded to an industrial computer for storage and processing. Statistical features of the electrical time series data are extracted using a sliding window approach and optimized with Principal Component Analysis (PCA). Subsequently, the binary tree support vector machine algorithm (BTSVM) optimized by k-means clustering algorithm is used to identify the detection states of the wood-based panels. Experimental results show that the system can achieve traceability of the formaldehyde detection stage records of artificial boards, and the designed algorithm model can also effectively identify the detection status of wood-based panels. The system and methods proposed in this paper have significant prospects for implementation in monitoring and traceability management within the testing industry.

The growing complexity of real-time video surveillance systems has made low latency and high stability critical challenges. Existing systems are hindered by redundant multi-node interactions, inefficient signaling, and rigid decoding strategies. While individual optimizations such as TLS reuse and adaptive buffering are established, their isolated application yields limited gains. This paper presents a systematic end-to- end optimization framework whose primary contribution lies not in individual techniques, but in a novel orchestration model for hierarchical video systems. This model is built on two core principles: strategic decoupling of signaling dependencies and a cross-layer adaptive control loop between network state and buffer management. By integrating metadata preloading, dynamic token authentication, SDP templating, asynchronous RTSP scheduling, and network-aware decoding, the framework achieves what isolated optimizations cannot: a 57.83%~65.85% reduction in end-to-end latency, 70% lower decoding waits, and sub-second video loading in practical deployments. The work establishes not only a scalable optimization paradigm but also a generalizable architectural blueprint for future distributed video systems.