基于改进的GSC-YOLOv7的多类型生产组件识别与定位方法OA

In view of the identification and positioning of multi-type components in the vision of the robotic arm,a lightweight neural network target identification and positioning method for auxiliary robotic arm grasping production components based on YOLOv7 target detection algorithm in the production process of intelligent factory is proposed.The improved GSC-YOLOv7 algorithm adopts the new network structure and lightweight neural network Ghost convolution module,and introduces the SE and CBAM attention mechanism to increase the weight of the required features,which solves the disadvantages of large number of parameters of the original neural network,large floating-point computation,low GPU utilization,and slow real-time detection speed.The experimental results show that:under the premise of balancing real-time detection speed and accuracy,the robotic arm vision system can meet the requirements of real-time detection and positioning;the real-time detection speed reaches 87.43 f/s,the training accuracy mAP@0.5 reaches 93.43%,which is increased by about 0.91%,and the detection accuracy reaches 97.94%;compared with the original YOLOv7 algorithm,the real-time detection speed is increased by 6.39%,the GPU utilization rate is reduced by 20.50%,and the training time and model volume are compressed by 27.63%and 46.77%respectively.