An improved YOLOv8 algorithm for indoor critical target detection
With the development of social service robots,indoor target detection has become an important task for robots to identify scenes.To address the low detection accuracy,slow detection speed,and difficulty in applying to embedded devices in the task of indoor target detection of existing networks,this paper proposes an improved lightweight network based on YOLOv8 algorithm.First,to overcome the difficulty in identifying scenes,a detection head is added to improve the detection accuracy of small targets.Then,Ghost Bottleneck is introduced to replace the bottleneck in the C2f module in the Neck part of the network and the SiLu activation function in the convolution in the latter half of the network is replaced with the H-swish activation function to reduce the number of parameters and computation and to improve the network ' s performance.The number of parameters and the amount of computation are reduced,the detection speed improved and the difficulty of network transplantation decreased.Next,MRLA attention mechanism is added in the Neck part to strengthen the connection between different layers,increase the feature extraction ability and improve the overall recognition accuracy.Our experimental results show on the indoor scene dataset,the improved algorithm improves the average accuracy by 3.6％ compared with the original one.The detection speed is 72 frame/s.Meanwhile,the number of network parameters is reduced by approximately 11％ compared with that of the original one,meeting the accuracy requirements and real-time performance of detection and outperforming the current mainstream algorithms.
国家科技期刊平台
NSTL
万方数据
