Optimization of Multi-Target Detection and Path Planning for Multi-Intelligent Body Collaborative Autonomous Systems Combined with Deep Q Learning

Deep reinforcement learning to help multi-intelligence to achieve multi-target detection of the environment and optimally plan paths.To this end, the article proposes an improved YOLOv7 method for image multi-target detection.Null convolution with mean pooling is introduced in the SPPCSPC module to shallow information in the image, and a lightweight SimAM attention mechanism is introduced in the head network to focus on the region of interest.And the hybrid edge regression loss function is proposed by combining NWDloss loss with CIOU loss.Meanwhile, the article further utilizes deep Q-learning for path planning on the basis of multi-target detection algorithm, and proposes two mechanisms to improve the DQN algorithm based on adaptive exploration strategy and based on changing the objective function for the two problems of DQN algorithm in path planning.In terms of datasets, compared with the YOLOv7 algorithm, the proposed algorithm improves the [email protected] of all detection categories by 3.2 percentage points, and is more than 3.1 percentage points higher than the [email protected] of the YOLOv7 algorithm, effectively realizing multi-target detection.The results of path planning simulation experiments show that the algorithm in this paper is able to plan good AGV paths, and the stability and convergence speed of the algorithm have been improved.