With the popularity of electronic devices, the number of malware has increased dramatically, posing a serious threat to the digital world. Accurately identifying malware has become a research focus. However, there are many difficulties in the research, such as insufficient algorithm generalization ability, unbalanced datasets, and long processing and identification times. To address these problems, this study proposes a malware detection framework (VBDN) based on a convolutional neural network (CNN). The framework incorporates data visualization, balanced adoption, data augmentation, and convolutional neural network techniques to achieve over 90% accuracy in classifying malware on all four open-source datasets. The experimental work has two other contributions: first, it not only focuses on the overall recognition effect of the algorithm during the research process, but also on the recognition effect of each category with the help of a confusion matrix, which provides useful information for cybersecurity personnel, researchers, and others to carry out subsequent targeted research. Secondly, the balanced approach adopted in this paper has the following advantages: no need to construct a new dataset, consumes fewer hardware resources, automatically evaluates the sampling weights, etc. Additionally, to enhance the generalization ability of the algorithm and alleviate the overfitting problem, this paper employs data augmentation techniques to improve the adopted method. By comparing with several state-of-the-art algorithms, it can be observed that the VBDN framework proposed in this paper achieves the desired results in time with acceptable accuracy.

With the advent of Industry 4.0 era, smart manufacturing is rapidly developing, which puts higher requirements on data acquisition (DAQ). At present, the enterprise’s equipment in the data acquisition exists in the problem of many types of equipment, equipment heterogeneity, customised development, high cost, and so on. To solve this problem, this paper proposes an innovative approach to address the challenges in data acquisition by combining You Only Look Once version 8 (YOLOv8) and Optical Character Recognition (OCR) technology. First, the data area to be collected is labelled, and then the YOLOv8 model is trained, and the model with the shortest recognition time and the highest efficiency is selected with comparable detection effects, to fully guarantee the real-time data collection; second, in order to guarantee the accuracy of the data collection, the OCR model is trained for the second time using the dataset, which improves the performance compared to the most effective PaddleOCR by 8 percentage points; at the same time, considering the aging and weak computational capacity of individual devices, in order to achieve generalisation, this study adopts a client/server architecture, deploying the computational load required for recognition to the server in order to achieve the generalisation, stability and reliable operation of the system; lastly, the scheme is tested on 16 types of 30 devices to complete the test. The experimental results show that this automated data collection scheme is an efficient and versatile data collection scheme that not only reduces manual intervention during data collection but also enables seamless integration of data from different sources.

With the rapid development of drone technology, aerial small target detection has become an important research direction in the field of computer vision. This paper proposes a lightweight object detection algorithm based on YOLOv11, YOLO-LiRa (Lightweight Intelligent Remote-sensing Algorithm), to address the issues of small size, complex background, and dense targets faced by small object detection in aerial images. Incorporating Monte Carlo attention mechanism and partial convolution into the C3K2 module enhances the extraction of small target features and reduces computational complexity. Referring to the lightweight design concept of MobileNetV3 to optimize the backbone network, combined with GSConv and VoVGSCSP modules, the multi-scale feature fusion capability of the neck network is enhanced. Moreover, the feature map resolution and detection performance were optimized using the DySample upsampling operator. Experiments on the publicly available AI-TOD datasets have shown that YOLO-LiRa achieves 0.27 on the mAP50-95 evaluation metric, reducing the parameter count by 24.4% and computational complexity by 12.5% compared to the Baseline model, while achieving a good balance between detection accuracy and speed. Compared with other mainstream object detection algorithms, YOLO-LiRa exhibits more competitive performance in small object detection tasks. The method proposed in this article is suitable for applications such as unmanned aerial vehicle monitoring, intelligent transportation, and agricultural monitoring that require high lightweight and real-time performance.

As the variety and quantity of goods in modern warehouse management continue to increase, optimizing space utilization and ensuring the safe and orderly storage of goods have become critical challenges. High-rise shelving systems are increasingly favored by enterprises, but long-term use, collisions with stacker cranes, and overloading can lead to structural deformation of the shelves. If these deformations are not detected and addressed in a timely manner, they may result in serious safety incidents and significant property damage. To address this issue, this study proposes a zero-shot shelf deformation detection method based on multimodal data fusion. The proposed approach integrates Micro-Electro-Mechanical Systems (MEMS) sensors and image data to establish a real-time monitoring and alert mechanism. Specifically, MEMS sensors are employed for real-time acquisition of shelf status, with threshold values set to trigger an initial alert mechanism. Simultaneously, cameras capture shelf images, and multiple You Only Look Once (YOLO) models are used to detect and classify critical components of the shelf, such as beams and columns. YOLOv11n is ultimately selected as the optimal model for detecting these structural elements. Based on the detected beams and columns, further feature extraction is performed, and the sensor data is fused with these features. A K-Means clustering algorithm is then applied to conduct the clustering analysis. To address the issue of a lack of negative samples in the dataset, the study employs oversampling techniques, including SMOTE, ADASYN, and Borderline-SMOTE, combined with machine learning models such as Random Forest and Gradient Boosting Decision Trees (GBDT). The experimental results demonstrate that both Random Forest and GBDT achieved precision, recall, and F1 scores exceeding 95%, confirming the effectiveness and accuracy of the proposed method in shelf deformation detection. The multimodal detection method proposed in this study not only improves the accuracy and real-time performance of shelf deformation detection but also provides strong technical support for the safety management of warehouse operations.

With the rapid development of internet applications such as video conferencing and online learning, along with the widespread adoption of bandwidth-intensive services, traditional wavelength-routed optical networks (WRONs) can no longer meet these diversified bandwidth demands. As an innovative solution, Elastic Optical Networks (EONs) dynamically allocate appropriate spectrum resources and configure corresponding modulation formats based on user requirements and traffic volumes, effectively overcoming the limitations of WRONs. However, EON transmissions must adhere to constraints including spectrum consistency, non-overlapping, and continuity. Prolonged network operation generates idle spectrum fragments that violate these constraints, resulting in spectrum resource wastage, aggravated network congestion, and degraded overall performance. To address these challenges, this study proposes the Subtree Routing Algorithm Based on Multicast Service (STMS) and the Spectrum Resource Allocation Algorithm for Multicast Services Based on Sub-Routing (SR-MSRA). The core concept of STMS involves grouping destination nodes of multicast sessions and splitting a single large light-tree into multiple subtrees, where each subtree adaptively selects optimal modulation formats according to real-time spectrum states, thereby significantly improving multicast service request success rates. Building upon STMS, SR-MSRA further enhances request success rates and mitigates spectrum fragmentation by decomposing multicast requests into sub-routes and introducing a spectrum window utilization metric to dynamically evaluate and select sub-routes. This strategy efficiently utilizes dispersed idle spectrum fragments while satisfying continuity constraints, substantially reducing multicast blocking probability. Additionally, we formulate an integer linear programming (ILP) model and design a heuristic algorithm. Extensive simulations on standard network topologies demonstrate that STMS reduces service blocking probability by 4.61\% compared to the conventional Multicast Light-Tree Protection Algorithm (MLPA). SR-MSRA exhibits superior performance in reducing blocking probability 2.43\% improvement over STMS, improving spectrum utilization, and alleviating fragmentation, proving its effectiveness in leveraging spectrum fragments. This research provides novel insights for high-speed flexible resource scheduling in EONs and offers theoretical and practical guidance for network operators to manage spectrum resources efficiently under diverse Quality of Service (QoS) requirements.