Research And Implementation Of Multi-Objective Recognition System Based On Deep Learning

With the rapid development of artificial intelligence,the development and advancement of advanced technologies in the field of robotics has made tremendous contributions to many industrial production and social services.The autonomy of mobile agents has become a key research hotspot.To achieve complete autonomy,mobile agents need to have the ability to perceive information about the surrounding environment,the ability to analyze dynamic environments,and the ability to judge and enforce risks.The speed of human visual judgment is very fast and the accuracy is also high,which allows humans to make correct judgments in an instant.In daily life,when humans take a look at the video,the scene in front of the scene or the image,they can get information about the position of the object and their categories,so that the fast and accurate recognition ability enables humans to complete complex visual tasks without excessive subconsciousness.Thinking,such as conscious long-term thinking when driving,can make accurate judgments based on the scene environment quickly and accurately.Therefore,for autonomous agents,it is especially important to have a fast and accurate multi-target recognition algorithm.A good algorithm model can enable the autonomous agent to quickly and accurately sense the surrounding environment while performing tasks,and combine the auxiliary sensors in the system.Make a correct judgment,therefore,for autonomous mobile agents,real-time and accuracy become an important indicator.To this end,this paper studies and implements multi-target recognition,and studies and compares several typical deep learning target recognition algorithms.It is applied to the autonomous navigation of mobile agents by improving network structure and algorithm optimization.The main research contents are as follows:(1)Taking the multi-objective recognition model as the research object,researching and analyzing the current mainstream target detection algorithms and network structure,comprehensively analyzing the advantages and disadvantages of each algorithm and network structure,and its practicability for the autonomous mobile agent overall system.(2)Optimize the end-to-end target detection YOLO model,improve the network structure,reduce the missed detection rate of multi-target target recognition in complex scenes,and improve the accuracy of small target recognition.(3)Study TensorRT technology,optimize the neural network through TensorRT,and combine the target platform parameters to optimize the deployment of the trained multi-target recognition model,greatly improve the inference and prediction speed of the multi-target recognition system,and improve System identification efficiency and system real-time.(4)Research Robot Operating System,build the intra-system communication method and distributed system through ROS platform,and realize the construction and deployment of the final multi-objective recognition system.Finally,through the analysis of experimental results,the multi-objective recognition model based on deep learning after optimization can effectively complete the real-time multi-target recognition and localization function,which has better recognition effect,and the inference speed has obvious improvement,which can meet the requirements of real-time recognition.