With the deep integration of cloud–edge collaboration and artificial intelligence, image-based intelligent perception has been widely applied in fields such as healthcare and marine science. However, the traditional “cloud-side training—cloud/edge inference” paradigm faces challenges including bandwidth pressure, inference latency, deployment difficulty, and accuracy degradation caused by limited edge computing power and imbalanced sample distribution. To address these issues, this paper proposes a “cloud–edge collaborative image recognition task offloading framework driven by federated learning for training–inference collaboration and resource optimization.” The framework focuses on three aspects: integrated training–inference, collaborative inference offloading, and federated aggregation weighting. First, a cloud–edge integrated training–inference task offloading model is constructed, formally characterizing the task–resource–data relationships. Using containerization and YAML orchestration, we implement an end-to-end workflow covering cloud training, image distribution, and edge deployment, and build an experimental platform based on Kubernetes/KubeEdge, which significantly reduces data transmission time and overall execution latency. Second, to tackle the problems of limited edge computing capacity and low confidence of lightweight models, we propose a collaborative inference migration strategy triggered by overload conditions and the upward transmission of low-confidence samples. This strategy is integrated as a plugin into the Kubernetes/KubeEdge/Sedna framework and is validated in pathological image analysis and marine fish recognition scenarios, demonstrating improvements in inference efficiency and recognition accuracy. Finally, to address the imbalance of local data samples in federated learning, we introduce a weighted aggregation method that simultaneously considers local model accuracy, stability, and sample size. This approach increases the contribution of high-quality local models in global aggregation. Experimental results show that the proposed method outperforms FedAvg under multiple imbalanced scenarios, effectively enhancing global model accuracy and robustness. Overall, the proposed framework realizes a closed loop between training and inference with collaborative optimization, providing a scalable engineering solution for large-scale, real-time, and high-accuracy cloud–edge image recognition applications.

Yizhou Meng. Cloud–Edge Collaborative Image Recognition Task Offloading: A Federated Learning–Driven Framework for Training–Inference Collaboration and Resource Optimization. International Journal of Big Data Intelligent Technology (2025), Vol. 6, Issue 2: 94-103. https://doi.org/10.38007/IJBDIT.2025.060210.

[1] Zhang C, Xie Y, Bai H, et al. A survey on federated learning[J]. Knowledge-Based Systems, 2021, 216: 106775.

[2] Li, W. (2025). Discussion on Using Blockchain Technology to Improve Audit Efficiency and Financial Transparency. Economics and Management Innovation, 2(4), 72-79.

[3] Chen H Y, Chao W L. On bridging generic and personalized federated learning for image classification[J]. arXiv preprint arXiv:2107.00778, 2021.

[4] Tang X, Wu X, Bao W. Intelligent Prediction-Inventory-Scheduling Closed-Loop Nearshore Supply Chain Decision System[J]. Advances in Management and Intelligent Technologies, 2025, 1(4).

[5] Yang D, Liu X. Collaborative Algorithm for User Trust and Data Security Based on Blockchain and Machine Learning[J]. Procedia Computer Science, 2025, 262: 757-765.

[6] Xu, H. (2025). Optimization of Packaging Procurement and Supplier Strategy in Global Supply Chain. European Journal of Business, Economics & Management, 1(3), 111-117.

[7] Huang, J. (2025). Balance Model of Resource Management and Customer Service Availability in Cloud Computing Platform. Economics and Management Innovation, 2(4), 39-45.

[8] Gao Y. Research on Risk Identification in Legal Due Diligence and Response Strategies in Cross border Mergers and Acquisitions Transactions [J]. Socio-Economic Statistics Research, 2025, 6(2): 71-78.

[9] Li W. Building a Credit Risk Data Management and Analysis System for Financial Markets Based on Blockchain Data Storage and Encryption Technology[C]//2025 3rd International Conference on Data Science and Network Security (ICDSNS). IEEE, 2025: 1-7.

[10] Jing, X. (2025). Research on the Application of Machine Learning in the Pricing of Cash Deposit Products. European Journal of Business, Economics & Management, 1(2), 150-157.

[11] Huang, J. (2025). Research on Resource Prediction and Load Balancing Strategies Based on Big Data in Cloud Computing Platform. Artificial Intelligence and Digital Technology, 2(1), 49-55.

[12] Truong T. The Research on the Application of Blockchain Technology in the Security of Digital Healthcare Data [J]. International Journal of Health and Pharmaceutical Medicine, 2025, 5(1): 32-42.

[13] Xu Q. Design and Future Trends of Intelligent Notification Systems in Enterprise-Level Applications[J]. Economics and Management Innovation, 2025, 2(3): 88-94.

[14] Wu X, Bao W. Research on the Design of a Blockchain Logistics Information Platform Based on Reputation Proof Consensus Algorithm[J]. Procedia Computer Science, 2025, 262: 973-981.

[15] Wu, H. (2025). The Commercialization Path of Large Language Models in Start-Ups. European Journal of Business, Economics & Management, 1(3), 38-44.