Institution profile

Zhejiang Sci-Tech University

Academic institutionasia · cn
Official website
Research library61linked papers
Opportunities0open roles
Selected work

Representative Papers

Comprehensive Performance Evaluation of YOLO11, YOLOv10, YOLOv9 and YOLOv8 on Detecting and Counting Fruitlet in Complex Orchard Environments

Jul 01, 2024

Accurate detection and counting of immature and young apples in complex orchard environments remains challenging due to occlusion, varying lighting, and dense foliage. Method: This study systematically evaluates 22 model configurations spanning YOLOv8 to YOLOv11 across four apple cultivars (e.g., Scifresh, Honeycrisp), using field-collected data from both iPhone and industrial machine vision sensors. Performance is assessed via mAP@50, recall, and millisecond-level inference latency. Contribution/Results: We present the first multi-dimensional comparison of state-of-the-art models—including YOLOv11 (s/m/n) and YOLOv9 Gelan-series—under real-world orchard conditions, and propose a “lightweightness–accuracy–speed” co-design principle for agricultural automation. Results show YOLOv11s and YOLOv9 Gelan-base achieve top-tier mAP@50 of 0.933 and 0.935, respectively; YOLOv11n attains ultra-low latency of 2.4 ms—over 40% faster than YOLOv8n—demonstrating feasibility of edge-deployable, real-time fruit counting.

2 citationsRead paper

Malicious Agent Skills in the Wild: A Large-Scale Security Empirical Study

Feb 06, 2026

This study addresses the critical gap in empirical research and labeled data concerning malicious behaviors in third-party large language model (LLM) agent skills. We present the first open-sourced, annotated dataset of malicious agent skills, derived from a large-scale behavioral analysis of 98,380 skills in community registries. Our investigation identifies 157 malicious skills containing 632 vulnerabilities, uncovering two dominant attack paradigms—data exfiltration and agent hijacking—and revealing sophisticated exploitation techniques targeting shadow features and platform hook systems. By integrating behavioral verification, vulnerability discovery, kill-chain modeling, and a responsible disclosure framework, our approach facilitated the removal of 93.6% of identified malicious skills within 30 days of disclosure, establishing essential infrastructure for advancing LLM agent security research.

1 citationsRead paper
Recent publications

Latest Papers

Implicit, Yet Impactful: Understanding Hidden Dependencies in Java Projects

Aug 17, 2026

This study addresses the security and maintenance risks posed by implicit dependencies in Java, treating them as a primary subject for systematic analysis. Through large-scale data mining and dependency graph analysis within the Maven ecosystem, we quantify their prevalence and impact. Results indicate that 34.12% of projects contain implicit dependencies, with 48% experiencing breaking changes due to version drift. Furthermore, 36 CVEs are directly invoked, and 30.28% of affected projects are exposed to known vulnerabilities. Based on these findings, this work identifies key contributing factors and proposes four mitigation strategies that significantly enhance the security and maintainability of open-source software supply chains, thereby filling a critical gap in empirical research on implicit dependency management.

0 citationsRead paper