In this paper, we study how to optimize the resource allocation and scheduling in art data recognition task by AI algorithm to improve the recognition efficiency and accuracy. This study first analyzes the main challenges faced by current art data recognition, including data diversity, limited processing resources and complexity of recognition tasks. Then this paper introduces the existing resource allocation and scheduling strategies, and points out their limitations. Then, this paper proposes a new automatic optimization framework based on Deep Reinforcement Learning (DRL), which aims to dynamically adjust resource allocation and optimize task scheduling strategies, thus improving the overall recognition performance. In the experimental phase, the performance of DRL algorithm, SVM(Support Vector Machine) algorithm, decision tree and random forest algorithm in art art data recognition task is evaluated through four experiments. In the benchmark performance comparison experiment, the area under the ROC (Receiver Operating Characteristic) curve of DRL reaches 0.95. In the resource utilization efficiency experiment, the average Central Processing Unit (CPU) usage of DRL is 45% and the memory usage is 30%. In dynamic load adaptability experiments, DRL shows shorter task response time and higher system throughput under different load conditions. In the final long-term stability and scalability experiments, DRL maintains 95% recognition accuracy after long-term operation, and achieves 25% processing speed improvement after increasing computing resources. From the above data conclusions, it can be concluded that DRL strategy performs significantly better than other recognition machine learning algorithms in art art data recognition tasks in terms of benchmark performance and resource utilization efficiency, and indirectly emphasizes the potential and advantages of DRL algorithm in processing complex recognition tasks.