An assembly part identification and localization method based on deep learning

Industry 4.0 and Made in China emphasize intelligent production and assembly, and there is the problem of low accuracy and slow efficiency of assembly parts identification and localization in machine tool machining and assembly process. Therefore, based on the lightweight network, attention and information fusion mechanism, a component identification and localization algorithm LAI YOLOv5 is proposed. Firstly, in the YOLOv5 network structure, the problems of large number of neural network parameters, large floating-point operations, high graphics memory usage and slow real-time detection speed are effectively solved by lightweight operations of the convolutional layer. Then the attention mechanism is introduced into the backbone network to increase the tendency of feature extraction and enhance the saliency of the detected object. Finally, the feature fusion network is added with the cross channel information fusion mechanism to enhance the feature detection capability. The experimental results show that compared with the original algorithm, in terms of model structure, the number of parameters and network layers of the improved LAI YOLOv5 algorithm are reduced by about 45.98% and 28.46%, respectively, and the amount of floating-point operations is reduced by about 55.82%, while the graphics memory usage is reduced by 15.51% and the length of the training time is reduced by about 32.27%. Meanwhile, the training accuracy reaches 96.80%, the training coverage reaches 95.01%, and the real-time detection efficiency is improved to 100.739 fps with a detection accuracy of 98.62%.