This paper reports the final-stage outcomes of a university-level doctoral English teaching reform project implemented at a science- and engineering-oriented university in China. The reform aimed to reconstruct doctoral English education by clarifying teaching objectives, strengthening academic competence training, and innovating pedagogical models in the context of artificial intelligence and global academic communication. Based on questionnaire surveys, interviews, classroom observation, teaching practice, and blended learning experiments, the project established a multidimensional teaching framework integrating academic writing, critical reading, conference presentation, interdisciplinary discussion, AI-assisted learning, and ideological reinforcement. The study demonstrates that the reformed model significantly improved students’ academic communication competence, classroom participation, autonomous learning ability, and international academic awareness. The project further argues that doctoral English courses should move beyond general language instruction and become an integrated platform for academic literacy cultivation, humanistic education, and national discourse competence.
This case study uses the "family tree" as the core carrier to construct a teaching path that integrates the Data Structures course with Chinese family genealogy culture. It explores the method of integrating ideological and political education (IPE) by extending from the engineering case of "social responsibility" to "patriotism and familial affection." The case first explains the correspondence between the tree structure of family genealogy and the tree structure in data structures, clarifying the relationship between the concepts of ancestral progenitors, ancestors, and descendants in genealogy, and the root, parent, and child nodes in a tree. Based on the data processing needs of family genealogy, the core operations such as tree creation, pre-order traversal, and node searching are taught, simultaneously introducing the family inheritance stories and deeds of ancestors recorded in traditional genealogies. The teaching incorporates the practice of building digital family genealogies, guiding students to recognize the social responsibility of technology in preserving cultural heritage. This leads to an understanding of the connection between family memory and national history, establishing the link between professional learning, cultural inheritance, and national responsibility, and creating a "profession with warmth and IPE with depth" classroom environment.
Artificial intelligence (AI) technologies are rapidly transforming higher education by offering new tools that support learning, academic writing, and problem solving. Despite growing interest in AI in education, empirical evidence on whether AI-related competencies improve students’ academic performance remains limited. This study examines the relationship between artificial intelligence knowledge, perceptions of AI usefulness, and learning efficiency among university students. Using cross-sectional survey data from 91 students, the study employs ordinary least squares (OLS) regression to analyze whether AI-related variables are associated with academic performance, measured by grade point average (GPA). The results indicate that AI knowledge has a positive and statistically significant effect on students’ academic performance, suggesting that students with greater familiarity with AI technologies tend to achieve higher GPA. In contrast, perceived usefulness of AI and perceived advantages of AI for teaching and learning do not show statistically significant relationships with academic outcomes. These findings suggest that AI literacy and competence may play a more important role in improving learning efficiency than positive perceptions of AI tools alone. The study highlights the importance of developing students’ AI literacy and integrating AI responsibly into higher education teaching and learning practices. The results provide practical insights for educators and policymakers seeking to enhance learning outcomes in an increasingly AI-driven educational environment.
This paper takes the "Intelligent Security System V1.0" project, in which the author participated as a core R&D member, as the research object, and systematically reviews the complete research-based learning process from project initiation, literature review, technology selection, system architecture design, module development, integration testing, to final deployment and documentation. Rather than focusing on the theoretical derivation or code-level implementation details of a particular face recognition algorithm, this paper emphasizes how, in the context of a complex system engineering project, an intelligent security system integrating edge computing and face recognition technology was accomplished through independent inquiry, team collaboration, resource integration, and continuous iterative optimization. Through this project, the author gained a deep understanding of the core value of research-based learning—the transition from passive reception to active knowledge construction—and mastered scientific methodologies in engineering practice, including problem decomposition, experimental design, performance tuning, version management, and documentation. This experience has laid a solid foundation for further academic study and professional development.
With the in-depth advancement of urbanization in China, the urban green area has gradually increased. The demand for its maintenance and management is also constantly rising. Lawn mowing, as a daily task with high repetition and high labor intensity, urgently needs an intelligent lawn mowing device to replace manual labor. Semantic segmentation technology, as the key technical basis of the perception system of intelligent lawn mowing robots, can accurately identify and distinguish different targets in the scene, and accurately define the boundaries of the operation area at the same time. However, traditional image semantic segmentation methods have problems such as a large number of model parameters, slow reasoning speed, and limited segmentation accuracy, which are difficult to meet the requirements of real-time operations. In response to the above problems, based on the DeepLabV3+ deep learning framework, this study proposes an improved segmentation model. By proposing the serial hollow space pyramid pooling and feature fusion module, the performance of the model is enhanced. And in view of the current situation where there is a lack of public lawn scene segmentation datasets, a dedicated segmentation dataset is constructed for the lawn scene.
The persistent digital divide has evolved from a binary of access (haves vs. have-nots) to a nuanced spectrum of skills, ranging from basic Information and Communication Technology (ICT) literacy to the high-level abstraction and problem-solving inherent in Computer Science (CS) education. This review paper synthesizes three decades of research across educational technology, sociology, and computer science pedagogy to critically examine the gap between ICT literacy and authentic CS education. We propose a novel, globally applicable framework, the Competency Progression Ladder (CPL), which delineates five transitional stages: (1) Access & Basic Operation, (2) ICT Literacy, (3) Computational Thinking, (4) CS Fundamentals, and (5) Authentic CS Practice. The review identifies persistent barriers: infrastructural inequities (global South), teacher content knowledge gaps (universal), and sociocultural biases (gender and race). We analyze intervention models from six countries (Estonia, Rwanda, India, Brazil, USA, Finland) to extract scalable best practices. The paper concludes with policy recommendations for moving beyond digital consumerism toward digital authorship and innovation, arguing that authentic CS education is not a luxury but a civil right in the 21st-century knowledge economy.
Industry-education integration represents the core direction of higher engineering education reform in the new era. As a key carrier connecting classroom teaching with industrial practice, the College Students' Innovation Training Project (CSITP) serves as an important platform for cultivating interdisciplinary engineering innovation talents. To address the prominent “industry-education disconnect” problems commonly observed in current CSITP implementation—including topics detached from real industrial needs, weak hands-on training components, and limited educational outcomes—this paper takes the CSITP “Grain Depot Fire Safety Early Warning IoT System Based on Deep Learning” as the research object to explore implementation pathways and models for engineering CSITP under the guidance of industry-education integration. By deeply aligning the project with China's national food security strategy and the practical pain points of fire safety management in the grain depot industry, a full-chain practice system of “industrial demand—scheme design—system development—on-site verification—achievement transformation” is constructed. Through approaches such as collaborative guidance by school-enterprise dual mentors, phased task-driven implementation, and blended teaching combining virtual simulation with physical practice, the team completed the development of a grain depot fire safety early warning system integrating multi-source perception, intelligent recognition, PID precise control, and remote linkage. The results demonstrate that this implementation model not only effectively addresses practical industrial challenges in grain depot fire safety, but also significantly enhances students' engineering practice capabilities, innovative thinking, and team collaboration competencies, providing a replicable and scalable practical case and theoretical reference for industry-education integration in similar engineering CSITPs.
In this paper, an improved convolutional neural network based on VGG16 network model is proposed to solve the problems of whether the film image is damaged by the traditional method and the difficulty and low accuracy of the damage type. Use migration learning to prevent over-fitting due to a small number of data sets; The fusion of cross entropy loss function and center loss function can optimize multiple targets at the same time to get more accurate results. In order to effectively prevent the gradient explosion and the gradient disappear and accelerate the training speed of the network; the improved network model is called the BN-VGG16 network model. The data sets of four kinds of damage types, including crack, dehumidification, particle and scratch, were constructed by means of image shooting, random rotation, stretching and migration. The improved BN-VGG16 network model is trained and tested on the data set. The experimental results show that the recognition accuracy of BN-VGG16 for film image damage category can reach 77%. Compared with other models, this model has obvious advantages in recognition accuracy.
With the explosive development of generative AI large language models represented by ChatGPT and DeepSeek, the field of academic education is undergoing unprecedented systemic transformation. This study aims to explore the dual impact of AI large language models as a "new intellectual medium" on traditional educational paradigms, academic ethics, and talent cultivation objectives. The research finds that, on the positive side, large models accelerate the universalization of educational resources and the improvement of teaching efficiency through personalized tutoring, interdisciplinary knowledge integration, and research assistance, breaking the limitations of traditional "standardized instruction" and promoting the transformation of education toward "human-machine collaborative" inquiry-based learning. However, on the negative side, the "black box" nature of large models and the inexplicability of generated content trigger academic integrity crises, weakening of students' critical thinking, and risks of knowledge dependence, while traditional outcome-oriented evaluation systems face challenges of obsolescence. The research further points out that the deep embedding of technology is forcing academic education to return to its ontological essence from "knowledge transmission" to "cognitive forging." Based on this, this paper proposes the construction of a "responsible AI education ecosystem," advocating for a balance between technological empowerment and educational essence through restructuring curriculum systems (strengthening prompt engineering and AI literacy), establishing dynamic academic integrity norms, and developing process-oriented evaluation mechanisms. The research shows that AI large models are not the terminators of education but catalysts. In the face of this cognitive revolution, academic education must embrace technological dividends while adhering to the subjectivity of human education, ultimately achieving a paradigm shift from "teaching knowledge" to "learning to question and innovate."
To address the latency, privacy, and performance issues of traditional centralized face recognition systems, edge computing offers an effective solution. This paper details my comprehensive involvement in developing an edge computing-based intelligent security face recognition system, a key project of the National College Students' Innovation and Entrepreneurship Training Program. Throughout the project lifecycle, I independently optimized an ECA-CNN-ViT hybrid model, constructed a three-layer cloud-edge-terminal architecture, and verified system performance. Beyond detailing the technical implementation and the resolution of hardware/software conflicts, this paper summarizes the practical engineering experience gained. It also reflects on the broader impacts of AI on computer science education and career planning, identifying personal areas for improvement to lay a solid foundation for future academic and professional development.
