Research on Key Technologies of Deep Learning Algorithms Enabling Robot Target Recognition and Intelligent Localization in Complex Environments

In complex environments, robot target recognition and intelligent localization face challenges such as multi-object, dynamic changes, and occlusion, making it difficult for traditional algorithms to meet the requirements of high accuracy and real-time performance. Based on this, the research improves the YOLOv5 algorithm to construct a target recognition model suitable for complex scenarios and optimizes intelligent localization capabilities by combining technologies such as MobileN etV3 and LSTM. Experimental results show that the improved YOLOv5 outperforms mainstream algorithms in various complex environments. In indoor occlusion scenarios, the average precision (AP) reaches 83.4% with an inference speed of 14.3 ms; in dynamic scenes, the AP is 76.7% with an inference speed maintained within 13.8 ms. Comparisons on the Cityscapes dataset indicate that the accuracy of the improved YOLOv5 ranges from 92% to 96%, surpassing Faster R-CNN. Under conditions of high noise and high-speed target movement in dynamic environments, its accuracy is 80.2%, which is approximately 5% higher than traditional algorithms. In summary, the proposed improved method in this research significantly enhances the performance of target recognition and intelligent localization, demonstrating robustness and adaptability in complex environments. It provides effective support for robot intelligent perception technology.