The integration of cloud computing, big data, artificial intelligence, and image recognition technologies is driving the intelligent upgrading of various industries. Cloud edge collaborative computing has become a key supporting technology for artificial intelligence image recognition by integrating the strong computing/storage capabilities of the cloud with the low latency real-time processing advantages of the edge. However, existing research has significant shortcomings in the integrated process of model training inference, joint inference strategies for edge resource constrained scenarios, and imbalanced samples in federated learning, which restrict its efficient application. To this end, this article focuses on three aspects of research in the field of artificial intelligence image recognition: firstly, a cloud edge collaborative image training and inference integrated task offloading model based on Kubernetes/Kubeedge framework is constructed, which realizes the full process automation deployment of cloud model training, mirror issuance, and edge inference; Secondly, a resource constrained cloud edge collaborative inference task offloading strategy is proposed, which triggers collaborative inference by monitoring edge load overruns and inference probability values; Finally, to address the issue of imbalanced samples in federated learning, a comprehensive weight evaluation method based on local model accuracy, stability, and sample size is proposed to optimize global model aggregation. The experimental results show that the integrated process can reduce data transmission delay and improve inference response speed. The joint inference strategy has significantly better inference efficiency and accuracy than traditional methods in medical pathology and marine fish image classification scenarios. The federated learning aggregation method effectively weakens the influence of sample differences and improves the accuracy of the global model in imbalanced sample scenarios. This research provides a reusable cloud side collaborative architecture design and performance optimization scheme for the vehicle intelligent analysis system, balancing low latency, high accuracy and privacy protection requirements. In the future, it will expand the task resource data collaboration mechanism of the Internet of Things scene, explore the reasoning mechanism of the complex scene of the industrial Internet, and optimize the accuracy of the global model by combining sample imbalance processing and aggregation methods, to promote the efficient application of cloud side collaborative computing in a wider range of scenarios.

Linghong Cheng. Architecture Design and Performance Optimization of Vehicle Intelligent Analysis System Based on Deep Learning and Edge Cloud Collaboration. International Journal of Multimedia Computing (2025), Vol. 6, Issue 1: 102-110. https://doi.org/10.38007/IJMC.2025.060110.

[1] Saillenfest A , Lemberger P .Nonlinear Concept Erasure: a Density Matching Approach[J].  2025.

[2] Cui, N. (2025). The Practical Application of Traffic Flow Forecasting and Capacity Analysis. Journal of Computer, Signal, and System Research, 2(5), 65-71.

[3] Wang Z J .Closed-form solution-based fuzzy utility vectors acquired from trapezoidal fuzzy pairwise comparison matrices using logarithmic quadratic programming for improving fuzzy AHP decision-making systems[J].Journal of Computational and Applied Mathematics, 2025, 468.DOI:10.1016/j.cam.2025.116647.

[4] Hao, L. (2025). Research on Perception and Control System of Small Autonomous Driving Vehicles. International Journal of Engineering Advances, 2(2), 48-54.

[5] 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.

[6] Pan, H. (2025). Development and Optimization of Social Network Systems on Machine Learning. European Journal of AI, Computing & Informatics, 1(2), 73-79.

[7] Yu B , Yuan H , Li H ,et al.Long-Short Chain-of-Thought Mixture Supervised Fine-Tuning Eliciting Efficient Reasoning in Large Language Models[J].  2025.

[8] 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.

[9] Cai, Y. (2025). Research on Positioning Technology of Smart Home Devices Based on Internet of Things. European Journal of AI, Computing & Informatics, 1(2), 80-86.

[10] Zillmann P , Garavaglia A , Schenk A M ,et al.Reducing channel interference with complex-valued asymmetrical weighted overlap and add filtering:US17805493;US202200017805493;US202217805493A;US202217805493[P].US12255851B2;US2025012255851B2;US12255851B2;US12255851[2025-08-20].

[11] Lu, C. (2025). The Application of Point Cloud Data Registration Algorithm Optimization in Smart City Infrastructure. European Journal of Engineering and Technologies, 1(1), 39-45.

[12] Madhavi S , Praveen R , Jagatheswari S ,et al.Hybrid ELECTRE and bipolar fuzzy PROMOTHEE-based packet dropping malicious node mitigation technique for improving QoS in WSNs[J].International Journal of Communication Systems, 2025, 38(2).DOI:10.1002/dac.5974.

[13] 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.

[14] 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).

[15] Zhu, Z. (2025). Application of Database Performance Optimization Technology in Large-Scale AI Infrastructure. European Journal of Engineering and Technologies, 1(1), 60-67.