Research on the Construction of Robot Target Recognition Systems Driven by Deep Learning Algorithms and the Optimization of Intelligent Localization Technology Performance

Robot target recognition and intelligent positioning technologies are widely applied in industrial production, logistics transportation, service robots, and other fields. However, they still face challenges in terms of stability in dynamic environments and accuracy in complex scenarios. To address these issues, this study designs an optimized algorithm that integrates multimodal data, combining a convolutional neural network (CNN) with a deep reinforcement learning (DRL) framework to construct an efficient target recognition and path planning system. To improve recognition accuracy, multi-scale feature fusion and an attention mechanism are introduced to enhance the model’s adaptability to small targets. In terms of positioning, an improved reward function and data fusion mechanism are employed to achieve better positioning accuracy and path smoothness. Experimental results show that the proposed method achieves target recognition accuracies of $\mathbf{9 1. 6 \%}, \mathbf{9 4. 5 \%}$, and $\mathbf{9 6. 8} \%$ for small, medium, and large targets, respectively, in complex backgrounds. Localization error analysis reveals that the mean absolute error for small targets is 0.058 meters, while it is only 0.027 meters for large targets in complex backgrounds. In path planning tasks, compared to the BP-CNN algorithm, the optimized algorithm improves path smoothness by $\mathbf{2 2. 3 \%}$, reduces response time by 15.8%, and achieves a real-time response rate of $\mathbf{9 3. 4 \%}$ in complex environments. In summary, the proposed method demonstrates strong target recognition and positioning capabilities in complex environments, providing robust support for the further development of intelligent robot technology. There is still room for improvement in the recognition and robustness optimization of small targets in complex scenarios.